DESCRIPTION: Point of View and "Camera-Eye" in selected films of Hou Hsiao-Hsien


Cinemetrics lab is the latest addition to our site, and is still work in progress. When finished, the lab is envisaged to offer 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 chart which looks a little like a star map. It is a scattergraph each dot on which represents a film available on our database. Select areas by dragging a rectangle to zoom in and see better different areas of film history. Once you have found your film on this map you will see how it relates to thousands of other films on the x-axis of time (past 111 years of film history) and on the y-axis of average shot lengths. What we intend to do is to add more tools to the lab in order to augment its statistics apparatus and enhance its means of data visualizations.

Year/ASL chart. Note the Log10 scale on the y-axis. It allows us to see all films in the very widely spread high ASL range.

WARNING: This page is not Internet Explorer friendly. IE might display this graph erroneously or very slowly. Please use a faster and less buggy browser (Firefox, Chrome, Opera, or Safari.)
This lab's films are shown in yellow.

List of films in this lab:

Puppet Master, The: (7) ASL 81.8
Millennium Mambo: (7) ASL 95.9
Flight of the Red Balloon: (7) ASL 76.5
Cafe Lumiere: (7) ASL 66
City of Sadness, A: (7) ASL 42.4
Three Times [Zuihao de shiguang]: (7) ASL 29.5


Cinemetrics Graphs of Five Selected Hou Hsiao-Hsien Films, 1989-2008

Edo Choi and Ian Jones 

A City of Sadness (1989):

A City of Sadness(6)   157 min 41.6 sec (as measured w/ credits sequences); 157m (officially listed running time)

223 shots (219 shots w/ 1 title card & credit sequences) for an ASL of 42.4 sec, SD of 33.5 sec

The Puppet Master (1993):

The Puppet Master(6)  139 min 8.1 sec (as measured, excl. opening/closing credits); 142 min (officially listed running time)

102 shots (100 shots, plus two title cards) for an ASL of 81.8 sec, SD 72.4 sec.

Yellow: Exterior - Extreme Long Shot, 27 shots for an ASL of 40.3 sec.

Orange: Very Long Shot, 16 shots for an ASL of 62.5 sec.

Red: Long Shot - Medium Long Shot, 28 shots for an ASL of 109.6 sec.

Purple: Medium Shot - Medium Close-Up, 5 shots for an ASL of 39.2 sec.

Blue: Variable Shot, 9 shots for an ASL of 118.9 sec.

Light Blue: Li Tien-Lu, 6 shots for an ASL of 150.3 sec.

Green: Stage (Puppetry or Opera), 8 shots for an ASL of 122.7 sec.

White: Other, 3 shots (2 titles & 1 Insert) for an ASL of 14.7 sec.

Millennium Mambo (2001):

Millennium Mambo: (6)  97 min, 61 shots for an ASL of 95.9 sec, SD 85.5 sec 

Percent of shots with major camera movement:  77%

ASL of shots with major camera movement:  116 sec

Percent  of shots with only minor reframing:  5%

ASL of shots with only minor reframing:  42.5 sec

Percent  of shots with completely static framing:  18%

ASL of shots with completely static framing:  27 sec


Café Lumière (2003):

Cafe Lumiere: (6)  99 min, 90 shots for an ASL of 66 sec, SD 56.6 sec 

Percent  of shots with major camera movement:  58%

ASL of shots with major camera movement:  57 sec

Percent  of shots with only minor reframing:  19%

ASL of shots with only minor reframing:  108 sec

Percent  of shots with completely static framing:  23%

ASL of shots with completely static framing:  29 sec

Flight of the Red Balloon (2008): 

Flight of the Red Balloon: (6)  109 min, 86 shots for an ASL of 76.5 sec, SD 83.9 sec 

Percent  of shots with major camera movement:  71%

ASL of shots with major camera movement:  86 sec

Percent  of shots with only minor reframing:  23%

ASL of shots with only minor reframing:  23 sec

Percent  of shots with completely static framing:  6%

ASL of shots with completely static framing:  91 sec

Visual and Temporal Aspects of Narration in Millennium Mambo
Ian Jones

            James Udden, in his 2007 essay “This Time He Moves!” pinpoints Hou Hsiao-Hsien’s film Good Men, Good Women (1995) as a moment in which the director’s approach to shooting began to veer away from the long-take, largely-static camera aesthetic he had originally pioneered, and that became a hallmark of the transnational “Asian minimalism” style that exploded into popularity following his success [1]. From the current vantage point, it is clear that Hou’s subsequent films, particularly those of the 2000s, which include Millennium Mambo (2001), Café Lumière (2003), Three Times (2005), and, most recently, Flight of the Red Balloon (2008), have represented a series of transformations and new approaches in Hou’s style. Moreover, whereas Hou’s trajectory from 1983 through 1993 displayed some easily-observable trends—towards both longer takes and towards more oblique, ellipsis-riddled storytelling—his films since 1995 have exhibited more stylistic diversity, and less of a clear progression. (Udden has delineated some of these older trends, as well as more recent developments, though data on ASL and percentage of shots with movement in them published in both “This Time He Moves” and an earlier essay, “Hou Hsiao-Hsien and the Question of a Chinese Style” [2].)

            This project attempts to move past the scattered idiosyncrasies of visual approach and narrational technique in Hou’s recent films, towards more concrete insights in the area of narration: in particular, its temporal aspects, and also what might be termed narrational “point of view” or the “character” of visual access. The film chosen for analysis is 2001’s Millennium Mambo. The film represents something of an ideal test case for several reasons. In visual terms, it presents much more of a leap from Hou’s precedent than Good Men, Good Women did. Additionally, the charge can be made that the film is, on the whole, unfocused—and this charge is not completely without merit. However, for the purposes of this project, this fault of the film actually works to make it easier to focus on the specifics of style, to isolate those sections of the film that do work to create interesting effects, to work with micro elements of Hou’s evolution, without becoming encumbered with the burden of producing some sort of theory of the film’s functioning as a whole. (I have also considered Hou’s recent trends in visual narration in an analysis of Flight of the Red Balloon, a film that I actually believe to be a success, and would rank among Hou’s best. As a compliment to the analysis at hand, portions of that study can be found here for the interested reader.)

Visual Narration: Terms of Analysis

How do the visual-access aspects of narration relate to what is being narrated? And what is the most precise and least misleading way to refer to this relation? Is it the “point of view” of narration? Certainly narration can have “properties,” but could one go so far as to say it possesses “character”? One of the most basic and yet overwhelming problems of tackling problems of visual narration is pinpointing a defensible way to frame the issue at hand.

One of the most immediately apparent dangers rests upon the fact that there are several glaringly wrong ways to discuss the issue at hand. There is the danger of applying wholesale anthropomorphism to the camera, ascribing to it the status of “observer” with identifiable character traits. There is also the danger of leaping to identify the techniques and structures of narration with the “point of view” of the director. On the other hand, there is also a danger of overcorrecting in an attempt to neutralize these two dangers. It is possible to completely erase from one’s language all references to “cameras” having a certain “character;” however, as much admirable restraint as this approach exhibits, it willfully neglects the fact that viewers can be, in many cases, cued towards some form of anthropomorphizing. Talented filmmakers can exploit the careful use of these cues. Instances in these cues occur should noted as such, forming another layer of the film’s quantifiable traits. The analyst’s strive towards neutral and precise language should not punish a film that expertly guides viewers to detect aspects of narration that seem to exhibit “character.”

A good place to start when tackling these problems is some of the major literature already written on the subject. George M. Wilson’s book Narration in Light: Studies in Cinematic Point of View (1986) displays an attractive approach towards the defining of its own subject. Wilson holds a resistance to poorly constructed conceptions of cinematic narration, while simultaneously avoiding too-forceful overcompensation. 

Wilson defines his book as a study of the ways in which narration “controls and limits the character of the visual access that we have to its fictional events” [3]. This phrase, “character of the visual access,” carefully avoids uncritical anthropomorphism and points towards a more neutral middle ground. The book attacks the conception, attributed to both editing theorist Karel Reisz and to V. I. Pudovkin, of cinematic visual narration as an “ideal” or “ubiquitous” observer [4]. Wilson offers some practical objections to such a viewpoint: “Despite what textbooks on editing such as Reisz’s often claim or seem to claim,” he writes, “we do not see a straight cut, even within a scene, as representing the phenomenology of a shift in a perceiver’s visual attention” [5]. However, Wilson also bases his objections on a more fundamental difference in his conception of cinematic visuals: We do not see visual narration as “images qua images,” he claims, but rather see “the fictional world through the surface of the screen” [6]. “A fictional world of the cinema,” Wilson writes, “is essentially a visible world and that which is presented in the imagery is meant to be the immediate, intersubjective visual appearance of the fictional objects and events” [7].

Wilson’s invocation of an “intersubjective world” might make gestures towards the foundation of a rich philosophy of cinematic representation (seemingly influenced by Husserlian phenomenology); however, Wilson is not clear, precise, or methodological enough throughout his book to provide a way forward for other writers to implement his ideas within analyses. There are other weaknesses, as well. In place of “narrators,” Wilson suggests that one speaks of the “implied (version of the) filmmaker” when discussing the how information is presented to the spectator, the type of view the spectator is afforded [8]. Wilson insists on this awkward formulation as a way of avoiding its equation with the actual filmmaker; however, the analyst does not receive much in return for the use of this cumbersome terminology. The phrase “implied (version of the) filmmaker” still indicates that narration be thought of in terms of human points of origin, rather than in terms of its own identifiable characteristics. In this discussion of Millennium Mambo, it seems senseless to refer to the character of visual access in the film as an “implied (version of) Hou Hsiao-Hsien.” Wilson’s account, then, is highly limiting.

By far the most expansive discussion of issues of the concept of cinematic “point of view” is Edward Branigan’s Point of View in the Cinema: A Theory of Narration and Subjectivity in Classical Film (1984). The text is valuable for its rich taxonomic rigor, which successfully pinpoints and criticizes nearly every conceivable pitfall of the “anthropomorphism of camera” variety. However, there are significant limitations of the text, attributable primarily to the manner in which author’s doctrinal intransigence restrains his explorations of certain cinematic techniques [9]. As a result, Branigan’s account presents a few helpful contributions, but there are also damaging oversights of his taxonomy that muddle rather than clarify the issue at hand. To help specify some of the terms of this project, it is helpful to note some of these problems.

Branigan proposes that narration be defined as “not a person or state of mind, but a linguistic and logical relationship posed by the text as a condition of its intelligibility” [10]. Accepting this definition, the “narrator” must then be defined as “a symbolic activity—the activity of narration,” which, rather identifiable as any person (i.e., the director), is “a role or function—a particular relationshipwith respect to thesymbolic process of the text” [11].

Branigan then draws up three categories, or, as he terms them, levels, of “narrator,” defined by their position within the diegesis. A narrator either has 1) no origin, 2) an origin located in non-diegetic time and space, or 3) an origin in the diegesis [12]. According to Branigan, the first-level activity of narration, that with no origin, is always present, as an underlying omniscient narration [13]. Sometimes the other levels of narration are co-present, superimposed upon and obscuring it, and sometimes these other levels are stripped away, leaving only the omniscient no-origin narration. (Borrowing a phrase from Thomas Nagel, one might call this level “the view from nowhere.”) Aside from having no origin and being omniscient, Branigan assigns three other characteristics to this “view from nowhere” narration: 1) its apparent authority is precisely derived from its lack of origin, 2) it seems to be only “the world,” and 3) it has the tendency to exhibit a “pure, camera-like objectivity” [14].

As for the other two possible modes of narrators, Branigan seems to limit the “origin in non-diegetic time and space” mode to voice-over narrators; after offering this single example he does not broach the topic again [15]. The way in which Branigan introduces the third level, “origin in diegesis,” initially makes it appear as if the level could contain many possible subsets, of which narration through the subjectivity of an on-screen character would only be one mode. “A character or several characters as well as other sorts of entities may function as an origin,” Branigan writes [16].

This “other sorts of entities” designation seems to offer ample room for many different possible properties of narrator. However, these possibilities are gradually pared away by Branigan throughout the remainder of his book. When a viewer is “confronted by an anomalous device” in narration, Branigan writes in his “Character Reflection and Projection” chapter, “one of the hypotheses we try out is a metaphorical application directed towards the nearest sentient agent, usually a character[17]. Are there other forms of sentient agents, other than characters, that could be the target of the viewer’s metaphorical application? Branigan disallows such options. “I will not consider,” he writes, “certain moments which some would call expressions of—comments by—the author, but rather that production of expressive detail given by the character himself” [18]. Branigan’s resistance to a sloppy attribution of every “anomalous device” to simply the voice of the “author” is understandable—this is one of the central problems of Wilson’s account. However, Branigan’s successive formulations leave the analyst little room to maneuver. Any element of narration is limited to the options of possessing “no origin,” in which case it must be omniscient and most always exhibits pure objectivity, of possessing an “origin located in non-diegetic time and space,” in which case it must be voice-over narration, or of possessing an “origin in the diegesis,” in which, apparently, must be a sentient agent that is “usually” a character.

What would the alternatives to these options be? One possible option would be narration that is not simplistically reducible to the “author’s” point of view, but also not aligned with any specific character’s subjectivity, nor any specific on-screen sentient agent. How does one describe such a narration, a narration that, at least in a certain sense, has “no origin,” but at the same time does not exhibit a pure “objectivity”?

It is doubtful that space could be carved out within Branigan’s taxonomy for such a form of narration. Although he makes an exception for characters, he holds suspicion of any other formulation of narration that allows the possibility of the attribution of a human-like judgment.  Branigan warns of the conceit of “point of view as attitude” while discussing André Bazin’s review of The Crime of M. Lange: Branigan writes that Bazin assumes that “camera movement at that moment [during an elaborate circular movement] becomes the objective correlative of the viewer’s disposition toward narrative and character” [19]. Branigan chastises this “point of view as attitude” approach, claiming that it rests upon the assumption that “the bearer of the ‘attitude’ is ‘like,’ or simply is, a real person who expresses himself/herself in a way which communicates with the viewer as a person” [20]. Branigan insists that the analyst’s task is not to impart “bearers” with “attitudes,” but rather to deal with only “codes of character and the ideology of the text” [21]

Branigan, however, offers no helpful proposal on how such a flawless bracketing can be carried out by the analyst. How such codes and ideologies can possibly exist in a vacuum, divorced from all background, evacuated of all frames of comprehension that might reference such categories as “bearer,” “character,” “attitude,” or even “person,” is not explicated by Branigan. Apparently, a description of such a code or such an ideology and its functioning is unnecessary; all that is required is the acceptance on faith that this is, in fact, how meaning is made by all texts and their human interlocutors in all situations: through the manipulation of abstract symbols that never referto (in any sense of the word, no matter how abstract) anything outside of their immaculate, apparently frictionless, self-enclosed system. The analyst who does not accept this on faith is forced to move on to a more promising schema.

David Bordwell, in his book Narration in the Fiction Film (1985), devotes much less space to specific discussion of narration-as-point-of-view than the other authors discussed, and yet his account of narration provides a sustainable middle-ground that offers many insights. Like Wilson, Bordwell voices skepticism towards the adequacy of the “ideal observer” model, but unlike Wilson he make no recourse to imprecise categories of world [22]. Bordwell would agree with Branigan that narration “is not a person or state of mind,” and yet this does not prevent him from applying practical labels such as “knowledgeability,” “self-consciousness” and “communicativeness” when discussing the characteristics of the access a viewer is afforded to the narrative [23].   Bordwell grounds these terms in the work of Meir Sternberg (whose focus on the primacy of “hypothesis” on the part of perceivers aligns nicely with Branigan’s own use of the term when describing viewer activity while negotiating cinematic point of view). His approach towards cognitivist accounts of the interaction between the spectator’s perceptual capacities and prior knowledge and experience is considerably more flexible than Branigan’s “codes”-centered account [24].

Of primary concern to our discussion of visual narration in Hou’s films is Bordwell’s account of what he terms the range and depth of a film narration’s “knowledgeablity.” Range refers to narration’s restrictedness; from omniscience to sharp alignment with a particular character’s knowledge [25]Depth, on the other hand, refers to degrees of subjectivity and objectivity [26]. Bordwell proposes that there exists a wide scale of potential values for both range and depth, and that the values of the two axes can be treated as independent variables, related to each other in any number of ways. Great objectivity in depth does not ensure omniscience of range; a film could in fact be restricted to one character’s knowledge but at the same time provide very few cues to that character’s psychological life. Unlike Branigan’s three levels of narrators, each of which are defined by fairly restrictive conditions, Bordwell allows for what amounts to free play on a plane of coordinates, with many unique but describable possibilities of narration. Adding Bordwell’s further properties of degree of “self-consciousness” and degree of “communicativeness” to the mix affords the analyst a range of simple values that can be used to describe a vast array of possibilities for the character of narration. Bordwell’s account of narration and the concrete textual properties of narrators will be a central touchstone of this analysis.

Preliminary Notes on Narration in Millennium Mambo

An immediately striking feature of Millennium Mambo is how deeply complementary its layers of narration are. In the film’s basic presentation of events, the narration tends to drift from the playful to the willfully beguiling. The film’s temporal structure has a tendency to drift into nonlinearity with unusual abandon; the rules guiding chronology in several of the film’s sequences are loose and seemingly ad-hoc. On a strictly visual level, the narration twists away from Hou’s previous maintaining of distance between camera and subjects to emphasize his baroque staging. Even in Good Men, Good Women or Flowers of Shanghai, which began the trend Udden writes on in “This Time He Moves,” the emphasis of Hou’s shots was largely still on actor’s bodies in relationship to each other and to their environment, with the camera acting as detached intermediary. In Millennium Mambo, the camera becomes much more of a self-consciously participatory agent. Elaborate, fluid tilts and pans accentuate even the most straightforward scenes. The film’s utilization of shallow depth of field, a new development in Hou, allows an exploration of isolated visual gradations, from figures in a distant landscape all the way down to macro-focused reflections. These visual trends can be simultaneously intoxicating and confusing, and represent a bold new direction of the “character of visual access” in Hou’s films.


Reflections, shallow depth of field creating multiple layers of focus, and camera movement as transformation and revelation in shot 11 of Millennium Mambo 

The claim that the film’s visual and temporal narration techniques not only compliment each other, but also comment upon the characters’ sense of rootlessness and lack of direction, might initially sound attractive, but is ultimately not very helpful. This is not a film that finds its anchor in its characters. To a certain extent, this is true of many of Hou’s films. On the other hand, Millennium Mambo may be fairly criticized as especially thin on characterization and lacking in serious depth. However, just because the narration of the film does not arrange itself in service of character, it does not follow that there is not an intimate relation between the two. Instead, Millennium Mambo feels particularly like a film in which characters exist simply in service of a certain approach to narration.

Consider the example of the film’s pre-credit sequence: in a slow motion traveling shot, the camera follows the film’s protagonist, Vicky, as she walks through a pedestrian bridge, occasionally glancing back to look directly at the camera. Voice-over narration intones: “She broke up with Hao-Hao, but he always tracked her down. Called her, begged her to come back, again and again. As if under a spell, or hypnotized, she couldn’t escape. She always came back.” 

What does this voice-over accomplish? Certainly, it sets up the character’s background, and the theme of her inability to progress: “again and again.” But rather than simply providing the viewer with exposition, the voice-over also has the effect of making much of what the viewer actually sees enacted upon the screen for large portions of time seem redundant: the viewer knows that Vicky’s fights with her boyfriend Hao-Hao will end with her return to him, “again and again.” And, indeed, that is the structure of the majority of the film. The narration establishes the character’s entrapment, while at the same time reflexively commenting on the film’s own tendency towards repetition-devices. But beyond simply commenting on itself, it also immediately works to set up a repetition device, right from the outset of the film: the “exposition” is not merely “exposition;” it is also something of a synopsis of much of the subsequent film. This is not simply self-conscious narration. It is narration that cannibalizes its own “content,” using narrative as a pretext for a self-enhancing feedback loop. Puzzles introduced in the first few seconds of the film remain unanswered throughout its remainder. Why is the voice-over narration narrating from the year 2011? Why narrate in the past tense from such a specific future point? Such a tactic cannot fit into the supposition that the narration is simply emulating the characters’ lack of direction; it instead sets forth a specific and unusual logic that will dictate the spectator’s relation to the events presented in the film. The pre-credit sequence functions more as a opportunity for the film’s narration to introduce itself than as a meaningful introduction to the character. (This is one of the aspects that is, admittedly, maddening about the film.) 

Narration and Time in Millennium Mambo

The notes above on the relationship between the past-tense voice-over and the events enacted on screen, and on the uses of repetition, redundancy, and nonlinearity in the film should have cued the reader in to the fact that any serious study of narration in Millennium Mambo should contain a section on the film’s creative uses of temporality.

When tackling this issue, it must of course be noted that the film’s temporal confusions are not unprecedented. Hou’s film The Puppet Master, in particular, is rife with ellipses—a technique that actually goes a long way in endowing the film with its power. However, the ellipses of that film, though they certainly have the potential to baffle, do not generally work to undermine the viewer’s overall conception of the linear progression of the chronology, only to deny context [27]Millennium Mambo takes a step towards denying chronology altogether.

One moment in the film is particularly striking both for the chronological confusion it creates in the spectator, and for the simplicity of the means by which it accomplishes this. Throughout the film, there exists a tension between the film’s voice-over narration, which falls into the category of what Bordwell would term recounting, “characters communicating information about prior events by any means,” and the on-screen action, which makes up Bordwell’s category of enacting, “when the syuzhet presents prior events as if they were occurring at the moment, in direct representation” [28]. This tension is muted in many instances, but fact that film’s voice-over narration switches between narrating events enacted on screen and events that are never enacted—all in the same past tense from ten years in the future—sometimes makes it difficult for the viewer to pick up on chronological markers in the on-screen events. Since relationship between the voice-over and enacted events is constantly in flux, this creates the potential for moments of strong disconnection.

Over the tail end of a shot in Hao-Hao and Vicky’s apartment (shot 9), the voice-over provides some exposition on Hao-Hao’s life and how he subsists without employment. This voice-over continues over a shot in an anonymous club (shot 10), where the paranoid and over-aggressive Hao-Hao gets into an altercation with another patron. The voice-over relates an incident in which Hao-Hao stole and pawned his father’s Rolex, with his father eventually catching on and calling the police to search his apartment. The on-screen action has no immediate connection with this anecdote, and at first it seems that the aural and visual narration have simply split, presenting two stories of Hao-Hao simultaneously, which add together to an overall portrait of the character—economic exposition. The temporal relationship between the events enacted on-screen and the events recounted on the soundtrack is indeterminate; however, at this point, this ambiguity presents no major difficulties to the spectator’s comprehension.

The film segues to Hao-Hao and Vicky in their apartment; however, once a succession of different views allows the space of the apartment becomes clear, it is revealed that this is, in fact, not the same apartment seen earlier, in shot 9 of the film. (This discrepancy is not explained; in fact Hou seems happy to willfully obscure this detail [29].) A few lines of dialogue between Vicky and Hao-Hao in shot 12 on the subject of the Rolex (“My father … the watch … he found out”) initially seem to function merely as a reestablishment of the on-screen action’s connection with the voice over narration. The sequence continues into a seemingly unrelated scene in which the couple’s hot water is shut off. (This progression is made more disjointed by its somewhat haphazard cutting; in this apartment location Hou introduces some jump cuts, a technique that he continues to use throughout the film, and also in such films as Café Lumière and Flight of the Red Balloon [30].) As Vicky waits around for hot water, the police suddenly arrive at the apartment, cuing the spectator to the fact that the Rolex incident is, in fact, still being enacted onscreen—and with alarming leisurely pacing, considering the viewer already knows the story’s outcome.

“The cops came. They found the pawn ticket,” the narrator blandly states at the 16-minute mark of the film; the police arrive on screen and proceed to search the apartment at the 24-minute mark. This search consumes three and a half minutes of utterly un-suspenseful screen time. Why this prolonged redundancy? Granted, this narrational device fits neatly into one of Bordwell’s schemas: story is recounted once, enacted once; it is recounted before we see it enacted. (This fits the definition of Bordwell’s “E” schema in his overall mapping of all possible relations between recounting and enacting of events in Narration and the Fiction Film [31].) And yet the voice-over narration seems so utterly detached: not simply free-floating, but placed so early and containing so much detail that it severely undercuts any possibility of tension in the upcoming scene. Instead of suspense, or even any remote feeling of curiosity as to what will happen next, the viewer is instead simply afforded only a sense of déjà-vu, of being stuck in a hazy state of repetition, a narrative path that contains too few markers, so that one does not realize one is traveling in circles until it’s too late. 

The “Dark Forest” of Yubari

Due to the “marked self-consciousness” prevalent in art cinema narration, Bordwell writes, viewer hypothesizing about character goals and their achievement is downplayed. The question of “what happens next?,” of primary importance in the narrative organization of classical Hollywood cinema, is displaced by a new central question: “Why tell the story in this way?” [32]Millennium Mambo, marked by long stretches seemingly devoid of character “goals,” constitutes a prime example of this type of narration. In previous sections, this analysis has attempted to tackle the question of “why tell the story in this way?” on something of a point-by-point basis. The question presents itself: is there any way to connect the film’s devices to some sort of large account or model, to piece together, or attach to, a broader theory or hypothesis?

As a starting point for this endeavor, I would like to turn to the film’s two sequences set in the village of Yubari, in Hokkaido, Japan. The village acts as something of an oasis for Vicky: she spends her time there far removed from the pressures of her usual social circle, and reverts to a state of unburdened playfulness unseen in much of the rest of the film. The two sequences themselves also present something of a narrative oasis for the spectator: they are meandering in the best possible sense; rather than feel unfocused and redundant, as large chunks of the film tend to, they feel expansive, and unburdened by time.

This account of the sequences and their function is, of course, highly subjective. To ground the discussion of these sequences, two attempts will now be launched: an attempt to explore, in concrete terms, exactly what separates these sequences stylistically from the rest of the film in terms of editing and visual narration, and an attempt to offer an explanation of their effectiveness in terms of a larger theory of time and narrative.

There are actually three sequences in the film in which Vicky travels to Japan. Two are set in Yubari, the third is set in the Shinjuku ward of Tokyo. In each of the three sequences, the average shot length drops precipitously compared to the film as a whole. This difference is noticeable in the Tokyo sequence—the ASL of this sequence is 66.7 seconds, with a standard deviation of 21.5 seconds, as opposed to the film’s overall ASL of 95.9 seconds (SD 85.5). However, the Yubari sequences go even farther: The first has an ASL of 34 seconds, and the second an ASL of 34.7 seconds, averaging together into an ASL of 34.7 (SD 25 seconds).  To put this into proper perspective: the overall ASL of Hou's films throughout the past decade is roughly 68 seconds.  At 95.9 seconds, Millennium Mambo is the slowest-cut film of this period by a good margin. The Yubari sequences of the film, by contrast, align more with the fastest-cut film of this period, Three Times, which has an ASL of 35.7 seconds [33]. Placed in an even broader perspective, these scenes resemble such roughly contemporaneous films as Michael Haneke’s Code Unknown (2000, ASL 35.6 sec) [34] or the Dardenne brothers’ Rosetta (1999, ASL 33.1 sec) [35] more than they do the rest of the film that surrounds them. Clearly, the Yubari scenes stand out from the rest of Millennium Mambo in terms of their editing.

      In addition to ASL compared to the Taipei scenes, the Yubari sequences are striking in their stylistic consistency with one another—they are so consistent that the second sequence feels like "reprise" of the first, despite its lack of actual overlap of narrative events. Both Yubari sequences open with vehicle-mounted shots (shots 23 and 55) on a snowy road to the small mining community. All vehicle-mounted shots in the film (shots 1, 21, 23, 44 and 55) are accompanied by the Lim Giong song "A Pure Person" on the soundtrack [36]. The use of this quietly propulsive track as a constant refrain endows these vehicle-mounted “transition” shots with a fleeting sense of directedness and momentum absent from much of the rest of the film. In vehicle-mounted shots in Taipei (for instance, shots 21 and 44) the track tends to fade out soon after the shot ends, quickly dissolving any sense of urgency the music may have briefly endowed the narrative with. By contrast, the song continues throughout the entirety of the first Yubari sequence, and throughout much of the second. (The second sequence crossfades the song into another track, which continues over the film’s end titles.)

      Following their respective opening shots, each sequence continues to follow a similar, though not identical, pattern. Each sequence is composed of seven shots. Following the vehicle-mounted lead-in shot, the first Yubari sequence composed of three shots with static framing and three in which the camera displays a mixture of panning and tilting. The second is composed of two shots with static framing and four in which the camera displays a mixture of pans and tilts. The statically framed shot, gradually diminishing in importance in Hou’s filmmaking, clusters in the Yubari sequences in Millennium Mambo. Static camera shots make up 36% of all shots set in Yubari (50%, if one considers that the vehicle-mounted shots in each sequence are in fact static in their framing, though they obviously reflect the movement of the vehicle itself), as opposed to only 18% of the film as a whole [37]. With the exception of one (quite short) shot in a club early on in the film (shot 7), static framing gains a foothold on the film and only finds regular use upon the first trip to Yubari.

      Beyond editing and camera dynamics, there is also the issue of color scheme, which overlaps with the nature of the setting itself. The visual difference between snowy, rural Japan and Taipei nightlife is, of course, quite striking. In Yubari, the color scheme tends towards soft blues and greens indoors, and wide expanses of blue-filtered snowy landscapes outdoors. This is a far cry from the harsh orange, greens, and odd blacklight-infused tones of Hao-Hao’s apartment, and from the deep blues and fluorescent highlights of the film’s club scenes. (The “Texound” club, introduced in shot 22, in which Vicky meets the Takeuchi brothers, who eventually invite her to Yubari, is a notable exception to the usual “club” color scheme of the film, lit with warm oranges and soft purples, and functions as both a narrative and chromatic transition zone from the harshness of Taipei to the comfort of Yubari.)


Contrast: the standard color scheme of Millennium Mambo’s club scenes, the “Texound” club, and Yubari

Color schemes and music alone can, of course, have profound effect upon a sequence’s emotional reception. The viewer’s overall association of the location with comfort, and with free-floating release from the repetitious grind of much of the rest of the film’s narrative is also, I believe, contributed to by the cutting rate and camera dynamics of the two sequences. The shot lengths in other locations—notably Hao-Hao’s apartments, with an ASL of 163.5 seconds—create a tendency towards tension that largely goes unresolved, creating a feedback loop of general unease. Hao-Hao’s physical examination of Vicky in their apartment in shot 9, for instance, is an uncomfortable scene, and becomes all the more so as the view refuses to cut away, subjecting us to its bare facts for 390 seconds—a length at which tension gives way to boredom, which gradually builds back into tension. An ASL of 34.4 seconds, while hardly fast, at least has the effect of breaking this cycle. The greater percentage of static shots in the sequences creates a similar effect. A camera style that emphasizes constant fluid pans and tilts within a long-take framework can, when employed for long enough without pause, begin to seem highly “probing,” and (quite literally) “restless.” Hou’s breaking of this pattern with an increase in statically framed shots in Yubari doesn’t simply add visual variety, it escapes a visual rut long enough to re-invest camera movement with a genuine sense of discovery rather than simply rote scanning of a familiar space. Ultimately, Vicky’s change in behavior upon arriving in Yubari cues the viewer to the place’s effect on her. The narration’s careful adjustment of its own parameters ensures that the place comes to have a similar associated effect in the viewer’s own experience.

A broader schema that I feel goes a good way towards explaining the power of the Yubari sequences in Millennium Mambo is presented by Paul Ricoeur in his essay “Narrative Time.” In “Narrative Time,” Ricoeur writes that, throughout the course of most narratives, readers “are pushed ahead” by narrative development and “reply to its impetus with expectations concerning the outcome and the completion of the entire process” [38]. However, this model is not without alternatives. One alternative is repetition. Repetition is, of course, a frequent device of Millennium Mambo, and one that often serves to draw attention to narration at the expense of the film’s narrative. But Ricoeur holds that there is another model of repetition, one in which circularity and linearity are synthesized, and an “existential deepening” of time occurs [39]. To illuminate this possibility, Ricoeur turns to the example of the “dark forest” trope of folktales. His exploration of this is worth quoting at length:

“Before projecting the hero forward for the sake of the quest, many tales send the hero or heroine into some dark forest where he or she goes astray…. These initial episodes do more than merely introduce the mischief that is to be suppressed; they bring the hero or heroine back into a primordial space and time that is more akin to the realm of dreams than to the sphere of action. Thanks to this preliminary disorientation, the linear chain of time is broken and the tale assumes an oneiric dimension that is more or less preserved alongside the heroic dimension of the quest” [40].

Here, Ricoeur has isolated, I believe, something more profound than simply what would be referred to in Formalist vocabulary as a “delay” or a “retardation device.” The Yubari sequences of Millennium Mambo do not “delay” some sort of goal-directed action of the narration and its characters; they transform the narrative by offering a genuine alternative to the goal-driven model. Ricoeur’s notes on the “realm of dreams” and “oneiric dimension” also seem highly relevant, given the ambiguous status of the return to Yubari at the film’s ending. Is it memory—a memory that just happens to have no overlap with the previously enacted events—or is it fantasy? Imagined future, or impossible return? Whereas the technique of redundancy throughout much of the film hints towards an aesthetic of boredom and directionlessness, and serves to heighten the sense of Vicky’s entrapment, both Yubar sequences, and especially the “Yubari reprise” present a strikingly different use of repetition, one that offers an escape into a realm far-removed from the frustrations of goals and deadlines.

The Effect of Location on Visual Narration

Millennium Mambo by location.  Yellow = Misc Ext Taipei, Orange = Club, Red = Hao-Hao’s Apartment(s),Pink = Yubari, Indigo = Kao’s Apartment, White = other

            Few of the other locations in the film display such easily isolatable visual/narrative trends when broken down for analysis. As has already been noted, shots in apartments shared between Vicky and Hao-Hao have the longest ASL in the film, at 163.3 seconds (SD 103.7 seconds). Shots set in Jack Kao’s apartment come in second, with an ASL of 127.8 seconds (SD 76.3 seconds), and shots in clubs come in third, with an ASL of 104.8 seconds (SD 80.4 seconds). Certain broad trends of stylistic variation in camera dynamics between locations are noticeable, but these fail to conclusively cohere into truly meaningful patterns displaying logics detached from the practical requirements of the settings themselves.  (My notes on this subject can be viewed here.)

On the whole, the “depth” of the knowledge of Millennium Mambo’s visual narration tends to skew towards objectivity, whereas the depth of its voice over narrator skews towards subjectivity. (It could perhaps be argued that the vehicle-mounted shots of the film tend towards subjectivity, since the music on their soundtrack coupled with their kinetic visual style seem to strive to express Vicky’s obvious euphoria in several of these transitional moments.) Both layers of narration remain strongly self-conscious throughout the film, as has been noted throughout this analysis. The “communicativeness” and the “range” of knowledge, on the other hand, are two characteristics of the film’s visual narration that tend to shift reliably with location.

The scenes in Jack Kao’s apartment, and the scenes in which in which he is present in general, tend to have the most expansive range of any in the film. Throughout most of the film, the narration’s access is restricted fairly closely to Vicky’s own experience. Several appearances of Kao provoke this restricted range to slip. Shots 19 and 20 take place in a sauna Kao is visiting, and although the voice-over narration is simultaneously explaining Kao’s entrance into Vicky’s life, it seems safe to assume that the moment enacted onscreen is Kao’s and Kao’s alone. Shot 44 is one of the film’s transitional vehicle-mounted shots, and shows Vicky riding in Kao’s car, however, unusually, Vicky is asleep. In shot 46, Vicky is passed out outside of the door of Kao’s apartment, and, in shot 48, Kao places a blanket on Vicky as she sleeps on his couch. Accentuating this broadening narrational range is the closed-circuit security monitor in Kao’s apartment, which adds a second, more static and impassive interlocutor-narrator between the shifting gaze of the camera and the characters themselves.

            By contrast, the scenes in the apartments Vicky shares with Hao-Hao display a range that tends to be aligned with Vicky’s knowledge, and also display a high degree of uncommunicativeness. It is in these locations that the narration is most prone towards temporal gaps (the frequent use of jump cuts, the odd navigation of the film’s overall timeline) and visual obfuscation (the inextricable layers of reflection in shot 11, plus the motif of vision obscured by hanging beads in shots 12, 14, 15, and 16). Especially notable in this regard is the relative lack of voice-over narration during the scenes set in these apartments. (Shot 9, the first shot of Vicky and Hao-Hao in their apartment, is the only exception to this rule.) Vicky’s misery under Hao-Hao’s thumb is largely explained to the viewer in voice-over dialogue that intrudes over scenes outside their apartments; when the two of them interact the viewer is usually denied contextual information, and forced instead to absorb a slow trickle of detail in the real-time duration of very long takes. This unusual mixture—arbitrary deletion of time through jump cuts, slow presentation of few narrative events through long takes, and frequent redundancy due to misalignment of scenes between Vicky and Hao-Hao and their respective voice-over explanation—works to create a sense of viewer frustration towards the narration that mirrors Vicky’s own emotions towards her situation (perhaps all too well).

            The narration’s range is most tightly restricted in the Tokyo scenes near the film’s conclusion, in which the viewer’s sharing of Vicky’s absolute ignorance of the fate of Kao creates both suspense and an odd aura of disconnectedness. The segue from this stifling isolation and unresolved suspense into the final Yubari sequence that closes the film is like the segue from drowsiness into the strange clarity of a dream.


            Hou’s style over the past decade has been much more diverse and less prone to identifiable trends than the decade-and-a-half that preceded it. Even when such categories as shot length and camera dynamics are analyzed systematically through a tool such as Cinemetrics, the bare data often cannot, on its own, indicate a logic behind his evolution as a filmmaker. However, this data can invaluably aid an investigation of narration in any given film.   This project has attempted to account for concrete developments in shooting and editing in Millennium Mambo through a broader theory of narrational point-of-view. Although Hou’s later oeuvre has lacked a sense of a gradual and purposeful shifting of parameters, analyses of his recent films reveal not that Hou’s style is scattershot, but that it is more heavily dictated by the logic of narration in any given film. A broad yet deep survey of Hou’s earlier and more recent work may serve to better illuminate trends in narration expanding far beyond the stylistic markers the director is most well-known for.


For some reason this lab continuously rejects my endnotes.  I have moved them here.  It might be best to open this link in a separate window or tab, to allow for easy comparison.