TAKING CINEMETRICS INTO THE DIGITAL AGE (2005-NOW)

by Yuri Tsivian

1. WHAT IS CINEMETRICS? The short answer to this already contains a digital component: Cinemetrics is an online application designed to simplify and amplify the kind of analysis described in David Bordwell’s brief survey and Barry Salt’s opening remarks to this section of Measurement Theory. In other words, Cinemetrics is an open-access interactive website designed to collect, store, and process statistical data about films. Its goal is to create an extensive multi-faceted collection of digital data which will enable researchers to explore structure and evolution of film editing. At the moment Cinemetrics is programmed to handle the aspect of editing known in film studies as cutting rates.

2. WHAT ARE CUTTING RATES? A peculiar thing about the film medium, noticed by many, is that it bridges the gap between spatial and temporal arts. On the one hand, filmmakers, like painters or architects, deal with recognizable spatial shapes; on the other, films unfold in time, as do poems or musical compositions. Though we tend to perceive their unfolding as continuous most films consist of segments called shots separated by instant breaks called cuts.

With rare exceptions, films contain a number of different shots. Shots differ in terms of space and in terms of time. We know enough about space-related distinctions between shots, which are easy to name (“shot 1: baby playing; shot 2: man looking”) and categorize (“shot 1: medium long high angle shot; shot 2: facial close up”). Time-related differences between shots are more elusive and harder to talk about, for, unlike in music or poetry with their scaled feet and measures, variations in shot length are not ones of distinction, but ones of degree. The only distinction a critic is safe to make when discussing shot lengths is between brief and lengthy.

Shot lengths are sometimes convenient to present as the frequency of shot changes, or cuts, hence the term “cutting rates.” Shorter shots lead to a higher cutting rate. Unsurprisingly, cutting rates are linked to the story and its space-time articulations: car chases are cut faster than park rambles, conversations shot in close-ups faster than medium-shots; likewise, “montage sequences” meant to cover larger spaces of story time will have higher cutting rates than will sequences shown in real time.

Less evident, but as important, is the relationship between cutting rates and the history of film; I will say a few things about this, for it is to find out about the historical evolution of editing that most people make use of Cinemetrics tools and browse through its database resources.   

3. WHAT FACTORS MAKE CUTTING RATES CHANGE ACROSS FILM HISTORY? We still do not know enough about this; to fill this gap in our knowledge we need to find ways to explore and compare statistically millions of shot-length data from cinema’s hundred-plus year history amassed at Cinemetrics database in seven years of its existence. What we know allows us to link changes in cutting rates to two facets of film history: film history as technology and industry, and film history as part of the history of culture. It is this triple task: to handle large quantity of data, to link them to hard facts about films, and to track their elusive associations with changing cultural taste that requires a teamwork by three specialists, not one.

It was due to technology, for instance, that the first films/shots produced by cinema’s French inventors Lumi?re brothers were all around 50 seconds each (for such was the capacity of their 1895 camera/projector), or that cutting rates jumped each time a new editing device was introduced in the more recent era – Scotch-tape splicing in the 1960s, editing on videotape in the 1980s or digital editing in 1994. But to explain why it was in the United States that the fast-paced “American cutting” was born in the 1910s, or how it happened that some 10 years later French and Soviet films managed to outstrip American cutting rates, one needs to address, as has been done, the state of the film industry: the specific mode of production then dominant in Hollywood, and, counter-intuitively, the non-dominance of this mode in post-WWI Europe.

Factors of style and culture further complicate the picture. Looking, for instance, at pre-revolutionary Russia with its taste for slow languorous film melodramas we find Russian film trade papers campaigning against “American cutting,” for here it was felt that “psychological” or pictorial acting styles – the main asset of Russian film stars – called for “full scenes” which must not be cut up. The 1917 Revolution turned the tables. Young Soviet directors like Sergei Eisenstein took over, declaring that the cinema of the future will need no actors at all – since anything an actor can convey will be much better communicated by means of cutting, or “montage.” It was this idea that fueled some of the fastest-cut pictures in the entire history of film, as well as well-known Soviet “montage theories” which claimed that the true constituent of the film is not the shot, but the cut.

4. AVERAGE SHOT LENGTHS. While debates about fast vs. slow cutting rates are central to the history of film, the notions of fast and slow will be of little use unless we have an idea of the normal. Distinct from the film critic, the student of film history cannot afford to rely on intuition, for as I have just shown the sense of cutting speed changed depending on when, where and by whom this or that film was made – saying nothing of different norms intrinsic to different genres. It is for this reason that an increasing number of film scholars resort to numeric data about cutting.

The method which film scholars interested in the history of cutting have been using for more than 30 years is based on calculating the Average Shot Length (ASL) of a film – an index obtained by dividing the length of the film in seconds by the number of shots in it. The result can be used in two ways. If we calculate ASLs for all the films made by the same director or edited by the same editor, and plot the results onto a timeline (diachronic statistics), we will get a better sense of their range of experimentation and creative evolution. Or we may choose to inspect a number of cross-sections of film history (synchronic statistics) and, by comparing their prevailing ASLs, get a sense of how cutting rates changed over the last hundred years.

It was the latter approach adopted by Barry Salt prior to 1992 and by David Bordwell prior to 2006 that yielded an overview of the way cutting rates fluctuate across film history. Having divided the span of film history into 5-year thick “splices” and calculated the mean ASL for each, Salt has shown the growth of cutting rates between 1912 and 1926, their decrease between 1928 and 1939, their relative stability during the forties and fifties, and their upsurge from the sixties to the eighties. And Bordwell’s more recent numbers show that between 1990 and now Hollywood films continue to pick up pace, the fastest of them reaching an ASL of less than 2 seconds.

I, too, once applied the ASL method in order to compare the last film made by the pre-Revolutionary Russian director Evgenii Bauer with the first film made by his Soviet successor Lev Kuleshov, and when I put the obtained ASLs side by side with the international data collected by others I felt my heart beat faster, for it turned out that between 1917 and 1918 the cutting tempo in Russia had jumped from being the slowest to being the fastest in the world. Not that the difference could not be sensed without counting, but I felt excited that now we could not only assume but also demonstrate this.

And it is not film scholars alone who are interested in determining exactly how shot lengths ebb and flow. In February 2010 a group of neuroscientists led by Cornell psychologist James E. Cutting published a study that connects fluctuation in average shot length with the rise and fall pattern inherent in human mind, a discovery startling enough to have made it to the pages of The New York Times (www.nytimes.com/2010/03/02/science/02angi.html). While Cutting’s study is not uncontroversial (see polemics between Cutting and Salt on “Cinemetrics Studies” http://www.cinemetrics.lv/salt_on_cutting.php ) it shows the range of fields in which average shot length statistics can be relevant.

Yes, ASL data work, but we need to keep in mind that these data are relational. ASL is useful if the only thing we need to know is how long this or that average shot is as compared to ASL figures obtained for other films, but it says nothing about each film’s internal dynamics. It was only after November 2005 after the Cinemetrics website became available to Internet users that film scholars can get a sense of how the cutting rate oscillates within the duration of a film.

5. WHAT CINEMETRICS BRINGS TO THE STUDY OF CUTTING RATES. Cinemetrics enables us to obtain and present cutting-related data in a more flexible way than previously available. Rather than calculate average shot lengths arithmetically, Cinemetrics records and stores the time-span of each separate shot. Distinct from the arithmetical ASL, which is a single datum, Cinemetrics treats each film as a database of shots and highlights its individual features. Specifically, it tells us about a film’s cutting swing (standard deviations of shorter and longer shots from ASL), its cutting range (difference in seconds between the shortest and the longest shot of the film), and its dynamic profiles (polynomial trendlines that reflect fluctuations of shot lengths within the duration of the film).

6. INVENTORY. In its present form, Cinemetrics includes: a) a software toolkit used for data collecting (the “client tool”) and data processing (the “statistics tool”); b) a database for storing the obtained data; c) Cinemetrics Labs, which help the user measure and assess a film against other films whose data are present at the database; and d) accessories: a discussion board, news board, two supplementary databases, and the section you are currently visiting, Measurement Theory, namely, a library of texts pertaining to its subject.

7. HOW CINEMETRICS WORKS. The way in which Cinemetrics interacts with its website users can be called a “tools for data” policy. This interaction takes 5 steps. The user (1) downloads the client tool free of charge; (2) uses the tool to measure the cutting rates of a film of his or her choice; (3) submits the measurements to the site; (4) upon submission, Cinemetrics processes these data using its statistics tools, and (5) stores them as part of the Cinemetrics database making the data available to other users.

The client tool (http://www.cinemetrics.lv/cinhelp.php ) allows you to register cuts, and not cuts alone. The tool can also be used to tag shots by seven scales, from Big Close up to Long Shot, or customized to measure the frequency of anything the researcher is interested in, be it characters, locations or music tunes. After the film has been measured and its data submitted, Cinemetrics automatically processes them, displaying the resulting information as a) statistical values and b) statistical graphs.

8. CINEMETRICS DATABASE (http://www.cinemetrics.lv/database.php ) is a shared-use, open-submission repository of data collected by people who use the client tool and processed by the statistics tools. Its default sorting is alphabetic by film titles, but it can also be sorted by other parameters such as year, submitter’s name, submission date, simple vs. advanced mode of measuring, and by the film’s average shot length, median shot length, and standard deviation. By clicking on a film title the user can access the page that provides basic statistics and interactive graphs related to this film.

Presently the database counts tens of thousands titles submitted by more than a thousand film researchers from different countries. Every new submission is announced on the “News” board, which one may visit to get a sense of the growing rate of submissions. A unique feature of Cinemetrics is that by submitting your film measurement data, you receive an analytical picture in return.

Take, for instance, one film from our database, say, John Ford’s Western from 1939, Stagecoach: (7) ASL 8.8. The wavy sparkline on this link-tag gives a basic idea about the internal ebb and flow of Ford’s cutting speed. To learn more, go to http://www.cinemetrics.lv/movie.php?movie_ID=2823 . Choose a value in the box “Trendline degree” to make the red line more sensitive to the data flow, or try out “ranges” 10 or 20 in the “Moving average” box for a snakier blue line to see how shot lengths vary. And if we want to see, at a glance, the distribution of seven shots scales from Big Close Up to Long Shot in Stagecoach all we need is to select “yes” in the “Color code?” box. 

9. CINEMETRICS LAB is under construction, but its basic features are already operative. It is envisaged to offer the students of film history a range of analytical tools that will help them dissect, visualize, and compare film-related data. We started with a large-scale comparative map that looks a little like a star map. It is a scatter-graph, and each dot represents a film available on our database. If you find your film on this map you will instantly see how it relates to thousands of other films on the x-axis on time (hundred-plus years of film history) and on the y-axis of average shot lengths. What we intend to do with the help from skilled statisticians as Mike Baxter is to add more graphic tools to the lab in order to augment its statistics apparatus and enhance its means of data visualizations.

10. WHAT DOES CINEMETRICS ADD TO WHAT WE KNOW? It may sound a truism, but it is one worth repeating: in science as in scholarship, progress is measured not by new answers given to old questions, but by new questions put to old answers. What narrative factors make cutting rates change within the duration of a film? What correlations there are between staging and editing, between the scaleof a shot and its duration? These are questions that can be asked looking at peaks and valleys of time-series graphs Cinemetrics generates for each submitted film.

What happened to film editing tempo when continuity scripts were introduced at some of Hollywood studios in 1911? How did the cult of opera divas impact the tempo of Italian film melodrama in 1913? Did Soviet avant-garde disdain for classical ballet or Stanislavsky’s method acting reflect on the cutting style of Russian post-1917 films? What dialectics binds Hungarian and Romanian “slow cinema” much in fashion among present-day’s cinephiles and super-fast action films like the 2007 Shoot 'Em Up: (7)  with an ASL of 1.6 seconds? These and other questions like these arise as we change the optics of study from micro- to macroscopic, as this collaborative proposal envisages doing in the next three years. Such is the scale of film history map the three of us will scrutinize together.