tag:blogger.com,1999:blog-8929346053949579231.post8638845322859443549..comments2024-03-11T02:10:31.396-04:00Comments on Sapping Attention: Fundamental plot arcs, seen through multidimensional analysis of thousands of TV and movie scriptsBenhttp://www.blogger.com/profile/04856020368342677253noreply@blogger.comBlogger23125tag:blogger.com,1999:blog-8929346053949579231.post-53938365769695960862015-06-04T06:33:42.716-04:002015-06-04T06:33:42.716-04:00Hi, your post is fascinating, thanks for sharing i...Hi, your post is fascinating, thanks for sharing it!<br />I am a master student and a newbie on topic modeling, but I have one question. Is the topic model trained to recognize a specific set of topics or is it trained only on script chunks (regardless their content)? I find LDA interesting but I would like to understand how to use it properly in this kind of analysis.<br />Thanks again!gianluhttps://www.blogger.com/profile/17256125477041711420noreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-62957023285341363502015-01-20T10:04:08.591-05:002015-01-20T10:04:08.591-05:00I wanted to end with a question. Like many others...I wanted to end with a question. Like many others, I think we're living in a golden age of television, something like a post-Wire era. One thing that makes the Wire and other shows like it so satisfying is that they have a story arc that stretches over a whole season, or even multiple seasons, rather than a single episode. In a real sense they are not episodic; the hour long episode is merely the delivery unit; the season is the unit of composition and experience, narrative development and closure. Each episode is more like a chapter in a Victorian novel than a short story featuring a known cast of characters. The Wire isn't on your big chart, but other arguably developmental shows like Mad Men or Weeds are. Even still, it's hard looking at episode averages to get a sense of whether longer, multi-episode story arcs register in your graph. Do they? If you plot all the episodes of developmental shows like Breaking Bad, would it form a larger arc structure, vs. episodic shows like CSI or MacGyver, which presumably retread the same space? Could your arcs, in other words, give us a way to measure how episodic or developmental a show is? <br /><br />Thanks again for your post, and for the ensuing conversation! dashorehttps://www.blogger.com/profile/10680264838076089896noreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-49302146247285147222015-01-20T10:02:49.265-05:002015-01-20T10:02:49.265-05:00In my first post I said that voice leading is homo...In my first post I said that voice leading is homogeneous whereas your procedure is heterogeneous. You proposed instead a metaphor of proximity: your arcs are "further" from the texts they represent than Schenker's analyses were to the compositions they analyzed. I think that's right: my initial binary was too stark. And I'm more willing than you expect to accept the idea that "plots are in some sense built out of words." (For a long while it was the fashion for literary scholars to suppose that there's nothing outside of the prison house of language.) Plots probably are built out of words: just not the words that appear in your topics or even in the texts themselves. Or at least, we can't assume that the words that build plots will appear in texts in the way that the dominant invariably appears in a Hayden composition (on an accented beat, not just a passing note, no less, and almost always at the beginning and/or end of a "sentence").<br /><br />This is actually a pretty intuitive point. Journalists structure their opening and ending sentences using the concepts of lede and kicker. But those words almost never appear in the sentences they write. Likewise, Greek tragedians had some idea that they should have a scene of recognition towards the end of their plays. Even before Aristotle named these scenes, they might have conceived of them using the word "anagnorisis" (recognition). But with possible exceptions the word anagnorisis doesn't appear in those scenes of recognition. All but the most artless or self-consciously artful (Joss Wheedon?) TV writers establish standard character types like the "ridiculously average guy" (http://tvtropes.org/pmwiki/pmwiki.php/Main/RidiculouslyAverageGuy) or the "plucky girl" (http://tvtropes.org/pmwiki/pmwiki.php/Main/PluckyGirl) without using the words average or plucky. So yes, plots are made out of words, but only rarely the ones that appear in the text and therefore in topics or in PCA reductions of topics. <br /><br />Sorry for taking time to explain obvious stuff. What's more interesting to me are times where an absent word structures a text as what Stephen Booth calls "an undelivered pun." My favorite example is the Herbert poem "Love III," which I taught just last week: <br /><br />Love bade me welcome: yet my soul drew back,<br /> Guilty of dust and sin.<br />But quick-eyed Love, observing me grow slack<br /> From my first entrance in,<br />Drew nearer to me, sweetly questioning<br /> If I lacked anything.<br /><br />“A guest," I answered, “worthy to be here”:<br /> Love said, “You shall be he.”<br />“I, the unkind, ungrateful? Ah, my dear,<br /> I cannot look on thee.”<br />Love took my hand, and smiling did reply,<br /> “Who made the eyes but I?”<br /><br />“Truth, Lord; but I have marred them; let my shame<br /> Go where it doth deserve.”<br />“And know you not," says Love, “who bore the blame?”<br /> “My dear, then I will serve.”<br />“You must sit down," says Love, “and taste my meat.”<br /> So I did sit and eat.<br /><br />The absent structuring word is "host": both someone who welcomes a guest and the sacramental bread that is also the body of Christ. Love starts as the first kind of host and becomes the second. The whole poem is a pun on a word that doesn't appear in it. (A.J. Greimas and other structuralists love this sort of thing and identify fantastic instances of it in French poetry.) If topic modeling might generate signs that, as effects, point back to absent words like anagnorisis or climax, or lede or kicker, it hard to imagine it pointing us back in a legible way to a word like "host."<br /><br />Anyway, I'm sharing the notion of an absent structuring word as a roundabout way of making the following point about the proximity of an analysis. The elements of a Schenkerian analysis (the tonic, dominant, etc.) nearly always appear as a real (metrically prominent) element of the work itself. Only in rare cases do the words that structure a plot (or a poem, or a news story) appear in the text they structure. dashorehttps://www.blogger.com/profile/10680264838076089896noreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-69384207859603238292015-01-20T10:02:06.369-05:002015-01-20T10:02:06.369-05:00Now with your arcs we're in a much more diffic...Now with your arcs we're in a much more difficult situation. The axes don't designate anything so simple or transcendental as pitch and rhythm; rather they are (if I'm getting the details right) reductions of 127 dimensions, each defined by a bundle of words that appear together with some frequency, to two dimensions. Let's put aside the vexed relation btw lexis and semantics and just say that you're dealing with semantic space. How many dimensions does semantic space have? How are those dimensions organized? How is space divided, marked, given distinct zones within those dimensions? I have no idea. In the absence of given, intelligible axes like pitch and rhythm, I suppose you could reorient the axes so that they align with some identifiable semantic feature, but then you'd risk losing the arc pattern to gain intelligibility, and part of the thrill of the whole thing is the extent to the reduced axes are "acheiropoieton," made by no hand. Alternatively, you could accept that PC1 and PC2 are strictly unintelligible, but that's rather unappealing, since it means both, yes, there's a real structure, and no, we don't have existing concepts to explain it. Those are the antinomies - is there a middle way? There may be, and perhaps you've already named it: one "directional" trend, and one "cyclical." Your arcs show that directionality and cyclicality manifest along multiple, complex axes of semantic space. This may be the furthest that one can plausibly push intelligibility. I also think it's important to hold out the possibility that culture creates complex, identifiable patterns that culture does not have the resources to comprehend. You may be bumping up against that limit. dashorehttps://www.blogger.com/profile/10680264838076089896noreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-9261737793372253842015-01-20T09:59:25.860-05:002015-01-20T09:59:25.860-05:00Whew! Thanks for taking the time to formulate suc...Whew! Thanks for taking the time to formulate such an awesome reply! I hope it's okay if I also respond in parts, and piecemeal at that. Like you, I'm doing my best to recall details from the theory and composition courses I took as an undergrad music major - so 15 years back now. From the look of it, you recall more of the intricacies than I do, but your mention of Caplin brought back lots of memories. The kind of formal analysis Caplin exemplifies is more responsible for my current interest in syntax than anything else in my studies, which also reinforces my general dissatisfaction with purely lexical analyses of language: in the interest of counting words, methods like topic modeling discard syntax. The procedure isn't illegitimate - we're always sacrificing one level of detail in order to see patterns at a different level - it just happens to sacrifice the aspect of language I care about most (and the one that most exercises actual linguists, for that matter). So I do think your approach is more Schenker than Caplin, since it disregards the syntax at the level of the sentence ( i.e. the local harmonic progressions) in order to detect larger lexical patterns (the Ursatz, the arc in flattened topic space). The analogy is never direct or satisfactory, but to my mind modeling topics is a bit like measuring the frequency of collocation of I V64-53, vi, IV, etc. within some segment of a sonata. It may tell you what tonal space you're in, but it necessarily ignores whether in their sequence the chords compose a perfect cadence, a false cadence, a half cadence, etc. But maybe this analogy just reveals, which is to say repeats, my particular stake in syntax against the isolation of lexis, which you are more than free to disregard. <br /><br />I do think that comparisons to other, established modes of analysis (whether in the same medium, as with narratology, or a different medium like music) are crucial for more than our own nostalgic reasons: they help to understand the stakes and procedures of any new mode of analysis like the "arceology" (nice!) that you've detailed in your post. Comparisons tell us what we're up to. <br /><br />One place to push further on the comparison is dimensionality. I don't know anything about Dmitri Tymoczko's approach beyond what I could glean from his website, but music has been understood mathematically and geometrically since Pythagoras at least. Setting aside the string or column of air that needs to vibrate in space, music divides the single axis of time at two different scales: on a very rapid scale (hundreds or thousands of oscillations per second), i.e. the frequency of pitches, and at much slower scale (tempo as measured in beats per minute), i.e. rhythm. We experience these ways of dividing time on different scales as two independent axes, and conventional western music notation depicts them that way: pitch on the vertical axis, rhythmic time on the horizontal axis. (I'm leaving out other potential axes of differentiation like instrumentation.) In tonal music both pitch and rhythm are largely governed by ratios of whole numbers (I know this isn't completely true, as the history of tuning shows, but it's basic nonetheless.) A whole measure gets split into quarters, eighths, sixteenths, triplets, etc. A fundamental frequency gets fractioned into partials by whole number ratios. So with tonal analysis you know both what the axes of possible reduction are and, more or less, the underlying system of ratios that differentiates rhythms and pitches along their respective axes. I wouldn't say that these axes are more "real" than principal components (both are real) but they are intuitive and necessary: to get transcendental about it, they are Kantian forms of intuition, the condition of experiencing any music whatsoever. At any rate, it is along the axes of pitch and rhythmic time that music is experienced, and along those same axes that different kinds of analyses and reductions occur.<br />dashorehttps://www.blogger.com/profile/10680264838076089896noreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-30079702954894877852015-01-17T18:52:08.822-05:002015-01-17T18:52:08.822-05:00(part ii)
Both of those suggest that maybe this i...(part ii)<br /><br />Both of those suggest that maybe this is wrong musicology to talk about. Rather than the analogy being the common practice period, it should something more like the classical style. (Less exciting than grand theories of narrative, but also more useful in understanding individual works). My undergrad music theory seemed to be dominated by <a href="http://www.amazon.com/Classical-Form-Functions-Instrumental-Beethoven/dp/019514399X/ref=sr_1_11?s=books&ie=UTF8&qid=1421535815&sr=1-11&keywords=musical+forms" rel="nofollow">Caplin</a>, which offers a taxonomy of medium-sized forms (sentences, periods, etc) that are slightly less rigorously rooted in the core elements of voice-leading. (Though Schoenberg's Harmonielehre, Caplin's back-progenitor, has as I recall an enormous emphasis on voice-leading; am I creating a false distinction?).<br /><br />And in the classical style, the constitutive elements of structure aren't exactly the notes or the harmonies that are the same as the notes; they're something else more impressionistic, closer to plot points. I'm thinking now, for instance, of Charles Rosen's analysis of why the Waldstein Sonata goes to the mediant rather than the dominant for the second theme in the exposition. As I remember, it could be caricatured as "because four sharps is four times as dominant-y as one."<br /><br />Which isn't to say that musical structure isn't much more real than plot arceology. (Why have I not coined that phrase until now?) But particularly in the most highly constrained forms--sitcoms, commercials, Reuters articles, possibly hymns--I think that might suggest there are forms where musicology could offer a model of a less internalist language than voice leading.<br /><br />Finally, you are absolutely right that the really interesting questions are about the individual deviations from the overall arcs. (By which I mean the multidimensional paths through token/topic space, not necessarily the 2d representations I'm talking about here). I'm still stuck, though, on the prior question--if there anything there stable enough to really draw a comparison to? Some of what I wrote above, and has come out on Twitter, makes me not want to completely close the door on that possibility. But certainly, the spaces I'm outlining so far aren't going to be enough to help a reading of "Memento" (or whatever) in any humanistic way yet. The question is, I guess, whether there is any way to taxonomize formal practices analogous to the sonata or the rondo on this that are common enough to permit general comparison.Benhttps://www.blogger.com/profile/04856020368342677253noreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-49568570628210038122015-01-17T18:51:50.213-05:002015-01-17T18:51:50.213-05:00Thanks for such a thought-provoking comment. You&#...Thanks for such a thought-provoking comment. You've given me enough to say that I've discovered for the first time Blogger's limit on comment lengths. My musicology is all rusty and undergraduate, so I'll probably be wrong some spots about what I'm about to say. (On the other hand, this whole post was about narratology, which I'm really far less entitled to spout about). But let me give it a shot. <br /><br />In many ways I would feel much more comfortable about my own analysis if it were about words, not topics, because I could at least make an argument (one I know you'd disagree with, probably correctly) that plots are in some sense built out of words. But alas, it would be much more illegible, and that's the point. Even if it were legible, though, you are clearly correct that the space of tokens described by the algorithms here is much farther than the one experientially from authors than the one of notes is from that navigated by composers. On the shorter horizons of voice-leading, in music they are nearly one and the same; that's one of the things so interesting about Tymoczko's work. (And probably others--I only really know him because he taught at Princeton when I was a grad student).<br /><br />Things are further complicated by the fact that even where "tropes" or "plot points" do exist, they are far less pervasive than the universality of tonality as it existed in the common practice period, and far less rigorous. I think you're right to point to Schenker's taxonomy of deviations as the thing I'm really interested in here, or why I see him as more "long-form." And those deviations are themselves important forms: evidently (Lord, I had forgotten how audacious this stuff gets) sonata form are the type where you hang out on two and then go back to three before heading back on down. So there's some hope of salvage.<br /><br />To pursue the analogy with the best face of Schenkerian analysis or voice leading, I'd probably have to choose between two different statements:<br /><br />1. Each major genre ("romantic comedy," "cop show," etc.) is essentially its own "tonality," and episodes can draw from multiple Ursatzen or depart from the pattern altogether.<br />2. There really are shared elements that tend to be elaborated upon--a uniform narrative structure, or at least set of gravitational poles, that most narratives tend towards. (I suspect these will be more thematic--"reflecting on progress"--than topical--"a marriage").<br /><br />(continued)Benhttps://www.blogger.com/profile/04856020368342677253noreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-30866056778544142952015-01-15T07:44:12.395-05:002015-01-15T07:44:12.395-05:00Fantastic:
"the problem resembles musicolog...Fantastic:<br /> <br />"the problem resembles musicology... So I'm curious what anyone else thinks on this one in particular." <br /> <br />I'm intrigued by the comparison to Schenkerian analysis; actually excited just to see another "word person" who knows what Schenkerian analysis is (though the more general, successor notion of "voice leading analysis" might be sufficient).<br /> <br />One thing to say is that Schenker proposed a kind of symmetry between analysis and generation: the structures revealed by reduction were the structures that composers working in the framework of tonal harmony elaborate (through techniques like arpeggiation or stepwise motion) in the first place. The basic structures (the tonic triad, the descending line (3, 2, 1)) are not merely what are revealed by analysis, but also what, through elaboration, generate the music as we hear it. And Schenker had reason to believe in this kind of symmetry because he studied the same kinds of composition techniques (harmony and counterpoint) that composers like Bach and Beethoven had studied centuries earlier.<br /> <br />Voice leading analysis is not only, in its conception, symmetrical but homogeneous. The structures that it identifies are (except in some weird cases where it must posit them, but that aside) actual notes that occur in the music. So reductionist analysis takes place in the same medium as the object it reduces. You can *play* the product of an analysis on a piano. <br /> <br />As I think you acknowledge, we should be hesitant to treat your PCA analyses in the same way. Topic modeling (at least as Blei describes it) relies on the fiction that "topics" generate documents, but it's important that we neither forget that this is a methodological fiction nor let it slip into an assertion. Put differently: the tonic and the descending line may (at least plausibly) generate the surface structure of music through elaboration. But TV writers are not generating scripts out of probabilistic collocations of words (certainly not as they occur in set 3 or 6 min segments). Insofar as we conceive of TV writing as having generative principles (it probably has many of them) at all, there's no reason to think those principles are composed of words, much less collocations of words - they may be "plot-points" or "tropes" or "situations" or "story elements. So your analysis is heterogeneous where Schenker's is homogeneous: topics may usefully correlate with some generative structure or another (the debate about whether this is "plot" or not is a useful one), but they are not those structures. This reminds me of Moretti's observation of the distinction between objects of history and objects of knowledge. The topics you identify are objects of knowledge but stand at a distance from the objects of history. <br /> <br />I think your post is pretty clear and right on about all of this - just wanted to help elaborate the comparison.<br /> <br />One more note about voice leading analysis: the point of its reductions (stripping away elaborations, simplifying rhythm, compressing prolongations, etc.) is not to "discover" the same hidden structure in every score (in the way that archetype analysis gets its thrill from discovering "the hero" in different texts). Rather, identifying the Ursatz (or rather positing it - you're right to see it as a kind of tautology) makes it possible to analyze the variously employed techniques of elaboration that individuate different works. The analogy to your post: the arc structure you "discover" is less interesting than the subsequent analyses of deviations from arc structure that it permits. (Though again, I suppose these to be deviations from a statistical norm, not different elaborations of underlying structures, as in Schenker, though perhaps you could come up with a taxonomy or classification of deviations as he did with elaborations.)dashorehttps://www.blogger.com/profile/10680264838076089896noreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-91626619379685805292015-01-07T10:23:55.657-05:002015-01-07T10:23:55.657-05:00Hi Erez! Yes, I worry about the classifier a bit. ...Hi Erez! Yes, I worry about the classifier a bit. Still, I think there's a chance even the strongest version of the Star Trek thing might hold. If typical episodes of TV shows tend to deal with big philosophical questions just as the beginning and end, and Star Trek does the same but deals with much bigger questions and deals with them all the time, then this might be a fair characterization. But I agree that it's unlikely that a first-pass method like this is actually clocking into exactly those features--what disturbs me the most is that the second-to-last chart shows some very strange patterns, where the arcs would look better if distorted to be rotated around a focal point. Something's fishy, and I'm going to pull away from PCA after about one more post.<br /><br />But I don't think I'm saying that "Star Trek" starts at the end, exactly: I'm saying it starts at the end compared to other plots. Say we were looking instead a metric for identifying laugh lines, and I found that action movies started off really jokey and gradually got less humorous. Then that method was applied to comedies, and found that they ended at a well more beginning-y place than the action movies began, and started at a far laughier place. That wouldn't show a problem with the classifier, because laughter-ratios actually are connected to genre; it would just be a generally interesting finding that both genres got less funny as they progressed. What's important are not the points in multidimensional space, but the directionality of movement among them. So PCA as a "classifier" doesn't do a great job on Star Trek--but the higher-dimensional classifier that tags it as "plotty" does work fairly well, because it moves through vector space from its beginning in a relatively predictable way.<br /><br />But of course you and Ted are right that it may be disingenuous to call this "plot." "Structural arrangement of thematic elements" would be something more like what's actually going. Benhttps://www.blogger.com/profile/04856020368342677253noreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-65484883891619312852015-01-06T15:48:46.841-05:002015-01-06T15:48:46.841-05:00About "the reasonable sniff test for plot str...About "the reasonable sniff test for plot structure" ... I just want to suggest that finding a digital proxy for "what critics mean by plot" might be a bad Grail to set off in quest of.<br /><br />First, because it would require very ambitious NLP. For instance, there is actually no rule that a novel has to have a single plot. Critics often talk about multiple interlacing plots. There could also be different narrators ... and pretty soon we are talking about a problem that is so to speak AI-complete.<br /><br />The other reason I think it's a bad Grail is that critics may not actually have a very well-formed conception of plot in the first place. I mean, we're still arguing about it.<br /><br />So I think the path to take here is probably to acknowledge that you're seeing linguistic and topical contrasts between the beginnings, middles, and ends of works -- things that are related to, but not the same thing as, our received (murky) understanding of the concept “plot.”<br /><br />The problem Erez is noting would still be meaningful then, but it becomes merely a question of choosing a reference point. When we say "beginnings and ends are less far apart in scifi," we could mean that a classifier <em>trained on SF</em> finds it harder to distinguish them than a classifier <em>trained on detective fiction</em> finds it to distinguish beginnings and ends in detective fic. Or we could mean that they are less far apart by a metric trained on the whole collection. Both questions might be meaningful (though I agree w/ Erez that the first formulation is probably what we're more likely to be asking). Neither question is quite what we thought we meant by "plot," but imho, that's okay.Ted Underwoodhttps://www.blogger.com/profile/04012428899328561750noreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-48411259150352383512015-01-04T16:47:36.613-05:002015-01-04T16:47:36.613-05:00Hey Ben -
This is really, really neat. I enjoyed ...Hey Ben -<br /><br />This is really, really neat. I enjoyed this whole post and think there is a lot to what you've uncovered.<br /><br />But I do worry that the fact that whole genre-s are more 'ending-y' is not so much a discovery as an issue with the method, an indication that genre features need to be controlled for more systematically. <br /><br />Seems to me that progressing faster-or-slower through the plot is something that can vary by genre, but that any method which effectively says s/t like: well this genre starts at the end and keeps going is no longer doing something that passes the "reasonable" sniff test for plot structure. <br /><br />I realize that what you are saying in the text of the blog post is more nuanced than this, but on first pass I think the data may simply indicate that star-trek is full of words that are ending-y for other genres, and so the classifier doesn't do a very good job on star trek. And indeed on many other specific shows.<br /><br />Notwithstanding that it does seem to get meaningful results out of the aggregate of all shows. <br /><br />Very cool!Erez Aidenhttps://www.blogger.com/profile/08981726064348596983noreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-32991367765572313602015-01-01T16:28:43.521-05:002015-01-01T16:28:43.521-05:00This comment has been removed by a blog administrator.Jhon Staphenhttps://www.blogger.com/profile/09109650156172127414noreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-62703480779858311422014-12-20T11:06:32.268-05:002014-12-20T11:06:32.268-05:00Kurt Vonnegut did something like this for his thes...Kurt Vonnegut did something like this for his thesis at U of Chicago. The faculty laughed him out of the University. Good luck for everyone who enjoyed his later work in fiction.<br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-19508820018182059812014-12-18T16:08:28.838-05:002014-12-18T16:08:28.838-05:00Yep. It's also going to be an interesting ques...Yep. It's also going to be an interesting question how far these components of "plot" for television do or don't overlap with the analogous components for prose fiction.<br /><br />Intuitively I tend to agree with Ben that this particular plot space is going to turn out to be pretty specific to television. E.g. the "sorry" topic feels televisual; we might borrow a phrase from Seinfeld and call it the "hugs and learning" topic. (Speaking of which, I wonder about Seinfeld...)<br /><br />But. We could be wrong.Ted Underwoodhttps://www.blogger.com/profile/04012428899328561750noreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-86623357135406504812014-12-18T13:52:47.988-05:002014-12-18T13:52:47.988-05:00On the first paragraph, I think this actually a pr...On the first paragraph, I think this actually a pretty important point--I see a few people here complaining that we all know that stories follow three-act structures, but actually the modern television three-act structure is a highly historical artifact that seems to be reinforced through strong conscious work <i>in specific media</i> right now. (It's hard to plot specific films, but even the most banal Hollywood films won't be so formulaic as TV, I believe. Or at least, that's one of the routes that can be explored with methods like this, but are difficult to do otherwise.)Benhttps://www.blogger.com/profile/04856020368342677253noreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-68846080121677345502014-12-18T13:46:17.703-05:002014-12-18T13:46:17.703-05:00What follows is pasted in from an e-mail from a fr...What follows is pasted in from an e-mail from a friend that I want to respond to here:<br /><br />> [...] And believe it or not I actually buy his [Dan Harmon's]/your/Joseph Campbell's mythology stuff: when I read Campbell a long time ago I couldn't see the point of his superficial all-myths-say-the-same-thing structuralism, but as a barometer of today's highly formalized pop culture mythology it actually makes sense. <br />> Have I given you my long-winded paean before about how The State guys (David Wain, Michael Showalter, Michael Ian Black, etc.) are actually pretty great narratologists? Their shows are constantly commenting on how the consistency of modern narratives have warped our sense of what counts as significant and eventful. But those deconstructions are usually equivalent to close readings rather than radical structural changes, and I'll be if you ran Wet Hot American Summer, Reno 911, or Stella through the bookworm bot the plot arcs wouldn't surprise you. <br />> But that doesn't mean I'd agree with your more dismal suspicion that Schenkerian sketches of this kind are banal or tautological. Cognitive narratologists are really trying to figure out what kind of conventional narrative order the mind (in both the individual sense and the cultural one) needs or wants or uses — not as a salve but as a scaffolding that can support a lot of renovation/rethinking at different levels. <br />> Or to put it another way, we are rarely aware of how concepts of time or causation are narrative fabrications until they're pointed out to us. The conceptual blending guys, Fauconnier and Turner, have talked about how our concept of "punishment" relies on exactly the kind of narrative that you pinged in one of your earlier charts (feel/hate/sorry/hurt/wrong fault): this is maybe an obvious one because we are more aware of judgments about guilt and retribution being selective readings of a situation because they're rooted in legal systems which are overtly deliberative. But you're also drawing out a lot of nonobvious conventions that are important for highlighting how deep our mythologies run (came/knew/wanted/saw/day/remember took is one of my favorite topics, and the Survivor pseudo-coda at point 4 in your PCA chart is very interesting too). I have a pet theory about the meta-arcing of your mini arcs but I'll save that for another day.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-50514307386705884232014-12-18T13:21:28.054-05:002014-12-18T13:21:28.054-05:00The paragraph about Star Trek is especially wild. ...The paragraph about Star Trek is especially wild. Deep relationship between plot structure and genre?!?!? Structuralists should looooove this stuff.Ted Underwoodhttps://www.blogger.com/profile/04012428899328561750noreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-82035024713372670512014-12-18T12:42:44.174-05:002014-12-18T12:42:44.174-05:00Wheel, reinvented. Congratulations. Wheel, reinvented. Congratulations. Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-14308402075689715592014-12-18T12:38:19.976-05:002014-12-18T12:38:19.976-05:00This is really excellent stuff. Basic structural p...This is really excellent stuff. Basic structural principles of narrative. I kind of can't even summarize my response. Superimposing the arcs of different shows is an excellent viz idea.<br /><br />Obviously, going to want to do this with novels as well. One could group novels by author or genre to create patterns as strong as the ones you're getting from television serials.<br /><br />Whew! Awesome idea.Ted Underwoodhttps://www.blogger.com/profile/04012428899328561750noreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-65246231765616849572014-12-18T08:55:02.293-05:002014-12-18T08:55:02.293-05:00Never seen so much blood come out of a stone befor...Never seen so much blood come out of a stone before.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-20619895357447201472014-12-18T08:49:53.470-05:002014-12-18T08:49:53.470-05:00Thanks. You're right that it's close to a ...Thanks. You're right that it's close to a traditional three-act structure, so it's probably worth asking what some of the differences from that might be. One is proportions-- compared to golden age or modern Hollywood, the "middle" here is longer than it should be. (Does that make it more like longer five-act plays?) Another might be content, actually. If I were going to describe the three batches of words, I might call them something like 1. Calm 2. Action 3. Reflection. The amount of post-facto exposition built into some dramatic forms is really different. <br /><br />And then there's the whole detective show thing... You can make the Epstein quote apply to that, but it's obviously by analogy. (Actually, I think the quote applies better to sonata form in some ways than it does to either detective stories or to romantic comedies. What's interesting about this space is that it applies to all the different genres fairly well--though it's totally unclear to me right now whether that's based on any real similarities, or on just having enough data to wrap them all in, as in my phony plot example. Benhttps://www.blogger.com/profile/04856020368342677253noreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-73674123290012325402014-12-17T21:13:32.588-05:002014-12-17T21:13:32.588-05:00Unbelievably good! Got this recommended by the Bro...Unbelievably good! Got this recommended by the Browser and it didn't disappoint. Great work.mexico guyhttps://www.blogger.com/profile/02143753974089681353noreply@blogger.comtag:blogger.com,1999:blog-8929346053949579231.post-44364386887386600672014-12-16T12:56:53.422-05:002014-12-16T12:56:53.422-05:00Wow, this is great stuff. You have analytically un...Wow, this is great stuff. You have analytically uncovered the "three acts" theory of story structure. As Julius Epstein, who wrote Casablanca, said to Vincent Sassone who was taking a writing course: "You're wasting your money. I'll tell you how to write a screenplay in three sentences. Act I, get your guy up a tree. Act II, throw rocks at him. Act III, get your guy outta the tree." Basically, there's the setup, the complications and the resolution.<br /><br />In other words, the structure you are detecting is completely intentional. It's also pretty old. You should dump some Euripides or Aeschylus into your analyzer. It's probably cross cultural too.Kaleberghttps://www.blogger.com/profile/05283840743310507878noreply@blogger.com