Kuleshov Effect

​

What is the Kuleshov Effect?

The Kuleshov Effect, as counted here, is a difference in the emotion ratings of neutral faces when a fearful scene, rather than a neutral scene, is placed between two looks at the face.

Results clearly demonstrated the presence of a significant effect in terms of both valence and arousal in the fear condition only.

That sentence is the abstract of Marta Calbi, Katrin Heimann, Daniel Barratt, Francesca Siri, Maria A. Umiltà, and Vittorio Gallese, published on 4 October 2017. Valence here means how pleasant or unpleasant the face seemed. Arousal means how stirred or calm it seemed. The faces in that count were different people, each shown once, all chosen as neutral. The famous film anecdote is a different object: similar close-ups of one actor. Framing Effect is the same facts in two wordings. Contrast Effect is a judgment pulled by a neighboring item. Neither one is this face-scene-face sequence.

​

The Kuleshov Effect in 3 Depths

• Beginner: A quiet face can be read as grief or as hunger because of the shot beside it. The printed recollection names soup, a coffin, and a child with a bear. It does not name a counted audience.
• Practitioner: Write pleasantness, arousal, and the emotion label as three lines. In the 2017 count, fear moved the first two. Happiness moved the label more clearly than it moved those two scores.
• Advanced: The 1.29-point arousal gap and the 1.25-point drop in pleasantness are one comparison, fear against a neutral scene, on scores recentered to each viewer’s own average. They are not points on the original scales, and they are not a license to add them into one “more emotional” number.

​

Origin

The English paragraph read here is Ivor Montagu’s translation of V. I. Pudovkin. In the Internet Archive text of Film Technique and Film Acting, the paragraph begins “Kuleshov and I made an interesting experiment.” The next page header in that text is “ON FILM TECHNIQUE 141,” so the paragraph ends on the preceding page of that printing. This pass did not match that scan to a 1929 title page. Montagu’s 1929 English Film Technique is the translation’s first book form. The year of the showing itself is not in the paragraph. Pudovkin says he and Lev Kuleshov took several close-ups of the well-known Russian actor Mosjukhin from some existing film. That spelling is Montagu’s. The same actor is also written Mozzhukhin, and in French Mosjoukine. This page uses Mozzhukhin after this sentence. The close-ups were static, quiet, and similar, and “did not express any feeling at all.” They were joined to other pieces in three combinations. A plate of soup stood on a table, and it was “obvious and certain” that he was looking at it. A coffin held a dead woman. A little girl played with a funny toy bear. The audience had not been told the secret. Pudovkin says the result was terrific. Viewers pointed out “the heavy pensiveness of his mood over the forgotten soup,” were moved by “the deep sorrow with which he looked on the dead woman,” and admired “the light, happy smile with which he surveyed the girl at play.” He ends: “But we knew that in all three cases the face was exactly the same.” That paragraph prints no date, no city, no head count, and no questionnaire. “Several” similar close-ups and “exactly the same” face are both his words. The paragraph does not say one frame was printed three times. Calbi and colleagues, citing a 1970 Pudovkin edition and Barratt and colleagues 2016, retell the contexts as a little girl playing with a doll, a dead woman in a coffin, and a bowl of soup, and they label those contexts happy, sad, and hungry. The doll is not the toy bear. Hunger is not the word pensiveness. A classroom version that uses a bowl of soup, a corpse, and a beautiful woman, and that names hunger, grief, and desire, is a third telling. It is not the paragraph above. Lev Kuleshov’s years are printed by Calbi as 1899–1970. The abstract of Daniel Barratt, Anna Cabak Rédei, Åse Innes-Ker, and Joost van de Weijer, Perception 45 (2016), pages 847–874, says the original film footage is lost. This page did not re-read that article’s tables. Pudovkin’s claim that he and Kuleshov made the edit together stays as he wrote it.

​

Key Points

Four readings keep the recollection, the recentered scores, and the category labels apart. A smile in a memoir is not a mean. A fearful scene is not a happy one.

​

Applications

Write the middle shot, the scale, and whether the face itself changed before a cut is treated as evidence of what someone felt.

​

Case Study

On 4 October 2017, Frontiers in Psychology published Calbi and colleagues’ count from the University of Parma. The paper was received on 25 May 2017 and accepted on 13 September 2017. A grant from the Chiesi Foundation to Gallese supported the work. Twenty-eight Italian adults took part, 14 of them women. The mean age was 28.1 years, the standard deviation 4.7, and the range 22 to 40. All gave written consent. The university’s review board had approved the study, and the authors cite the Declaration of Helsinki. No one is described as having been dropped for a questionnaire score. The subscale means are not the face ratings, and they are not printed here. Each person watched 18 silent gray sequences, six for each kind of middle scene. A sequence lasted 6 seconds: the face for 1,500 milliseconds, the scene for 3,000 milliseconds, and the same person’s face again for 1,500 milliseconds. The scene was what the face was supposed to be looking at. Faces were drawn from 24 neutral Karolinska pictures already chosen by Barratt and colleagues, shown in three-quarter profile, with a slow zoom. Each identity appeared once for a given viewer. Nine sequences used women and nine used men. Happy scenes included puppies, kittens, or newborns. Fear scenes included spiders, snakes, a growling dog, or war. Neutral scenes were mostly city and country views. The pool held 16 scenes of each kind. The supplement that describes how those scenes were chosen was not re-read here. Viewers first rated pleasantness from −4 to +4 and arousal from 1 to 9, with no time limit. The analysis recentered those scores on each person’s own three-condition average. Happiness then averaged 0.76 for valence (SE 0.2) and −0.21 for arousal (SE 0.2). Against neutral, the happiness coefficient in the context model was 0.33 (SE 0.2, t = 1.4, p = 0.17). Post-hoc tests did not find a significant valence or arousal difference between happiness and neutral. The model as a whole had chi-square (2) = 143.68, p below 0.0001. It accounted for 10 percent of the score variance before random effects (R2m = 0.10; R2c = 0.21). The second block repeated the 18 sequences in a new order and asked for one of seven labels. Fear was the most chosen label after fear scenes, at 24.4 percent, with sadness close at 22.6. Happiness was the most chosen label after happy scenes, at 35.7 percent, with surprise at 20.2. After neutral scenes the table’s mode is sadness at 25 percent. The lesson is narrow. A fearful scene between two looks at a neutral face shifted both recentered ratings in this room. A pleasant scene shifted the chosen label further than it shifted those ratings. The 1920s recollection, the 1992 null report as Calbi summarizes it, and this table are three different objects.

​

Boundaries and Failure Modes

The happy scene is the first boundary inside the same paper. Pleasantness and arousal after puppies, kittens, or newborns did not differ significantly from the city and country scenes. The category “happiness” still drew 35.7 percent of the labels. A lean in the label is not the valence result. Calbi’s own discussion says the fear result, and not the happiness result, is where the dimensional effect appeared, and they suggest that a strongly stirring context shows the shift more clearly. That suggestion is their reading. It is not a second experiment. The second boundary is the design. Scores were recentered, so −1.01 is not a mark on the original −4 to +4 line. The raw means are not printed. Twenty-eight adults in Parma, Italian, with sound removed, are the sample. Each identity appeared once, so a viewer did not see one photograph under all three scenes. Dean Mobbs and colleagues, in Social Cognitive and Affective Neuroscience (2006), used a different task: this page read their method and not their rating figure. They scanned 17 volunteers and analyzed 14 (8 women; mean age 27.5, standard deviation 8.6), who were naïve to the premise. Faces included neutral expressions and 25 percent morphs toward happiness or fear. Subjects were told the expression was a response to the juxtaposed movie. Those movies were zoomed pictures. No 2006 rating mean is printed here. Barratt and colleagues’ abstract says 36 participants each saw 24 sequences across happiness, sadness, hunger, fear, desire, and a neutral control, and that eye movements did not show the expected differences between emotional conditions. Calbi, citing their pages 15–16, says sad contexts were the most negative and least aroused, and desire contexts the most positive and most aroused. Those cells were not re-read here. The common misuse is to treat Pudovkin’s “terrific” as a percentage, or to carry 1.29, −1.25, 24.4 percent, or 35.7 percent onto a broadcast, a classroom grade, or a child’s face in a home video. An auditory version is a further paper. Calbi cites Andreas M. Baranowski and Heiko Hecht, Perception 46 (2017), pages 624–631, and describes that design as different from these sequences. It was not re-read. No music percentage is on this page.

​

Common Misconceptions

Three familiar sentences are wider than the sources read here. The recollection, the label, and the recentered score each stop at a different place.

​

Related Concepts

The neighboring pages change a judgment by wording, by a nearby item, by a prior word, or by a single trait. This page’s counted change is a middle scene cut between two looks at a neutral face.

​

One-Line Takeaway