Friday, September 9, 2016

The efficient plots hypothesis

I'm pulling this discussion out of the comments thread on Scott Enderle's blog, because it's fun. This is the formal statement of what will forever be known as the efficient plot hypothesis for plot arceology. Noble prize in culturomics, here I come.

Monday, August 29, 2016

Language is biased. What should engineers do?

Word embedding models are kicking up some interesting debates at the confluence of ethics, semantics, computer science, and structuralism. Here I want to lay out some of the elements in one recent place that debate has been taking place inside computer science.

I've been chewing on this paper out of Princeton and Bath on bias and word embedding algorithms. (Link is to a blog post description that includes the draft). It stands in an interesting relation to this paper out of BU and Microsoft Research, which presents many similar findings but also a debiasing algorithm similar to (but better than) the one I'd used to find "gendered synonyms" in a gender-neutralized model. (I've since gotten a chance to talk in person to the second team, so I'm reflecting primarily on the first paper here).

Wednesday, July 20, 2016

Why Digital Humanists don't need to understand algorithms, but do need to understand transformations

Debates in the Digital Humanities 2016 is now online, and includes my contribution, "Do Digital Humanists Need to Understand Algorithms?" (As well as a pretty snazzy cover image…) In it I lay out distinction between transformations, which are about states of texts, and algorithms, which are about processes. Put briefly:
Put simply: digital humanists do not need to understand algorithms at all. They do need, however, to understand the transformations that algorithms attempt to bring about. If we do so, our practice will be more effective and more likely to be truly original.
It then moves into one case study; the Jockers-Swafford debate of 2015, large parts of which hung on whether the Fourier transform was a black box and how it its use as a smoothing device might be understood. It's like a lot of what's on this blog, only better thought and edited.

The transformation/algorithm distinction is not a completely firm one, but I have found it extremely useful in a lot of research and teaching problems I've approached over the last year. So in addition to advertising that article for your consumption/fall syllabi production, I wanted to take the occasion to put on github a tiny little germ of a project to provide one-page, transformation-oriented introductions to basic text-analysis concepts that came out of using this thinking for a workshop on text analysis at the NIH in Bethesda, and describe what's in it. I'd love for anyone else to use it, fork it, whatever.

Monday, July 18, 2016

Plot arceology 2016: emotion and tension

Some scientists came up with a list of the 6 core story types. On the surface, this is extremely similar to Matt Jockers's work from last year. Like Jockers, they use a method for disentangling plots that is based on sentiment analysis, justify it mostly with reference to Kurt Vonnegut, and choose a method for extracting ur-shapes that naturally but opaquely produces harmonic-shaped curves. (Jockers using the Fourier transform, and the authors here use SVD.) I started writing up some thoughts on this two weeks ago, stopped, and then got a media inquiry about the paper so thought I'd post my concerns here. These sort of ramp up from the basic but important (only about 40% of the texts they are using are actually fictional stories) to the big one that ties back into Jockers's original work; why use sentiment analysis at all? This leads back into a sort of defense of my method of topic trajectories for describing plots and some bigger requests for others working in the field.

Tuesday, July 5, 2016

Nature publishes flat-earth research paper

I usually keep my mouth shut in the face of the many hilarious errors that crop up in the burgeoning world of datasets for cultural analytics, but this one is too good to pass up. Nature has just published a dataset description paper that appears to devote several paragraphs to describing "center of population" calculations made on the basis of a flat earth.

Monday, May 30, 2016

Literary Dopplegängers and interestingness

I started this post with a few digital-humanities posturing paragraphs: if you want to read them, you'll encounter them eventually. But instead let me just get the point: here's a trite new category of analysis that wouldn't be possible without distant reading techniques that produces sometimes charmingly serendipitous results.

I'll call it dopplegänger books. A dopplegänger is, for any world-historically great work of literature, a book that shares many of the same themes, subjects, and language, but is comparatively obscure, not widely read, and--most likely--of surpassingly mediocre quality.

Edit: Ryan Cordell informs me privately and regretfully that I'm wrong in some of my conclusions here. I said, "I took a grand total of one English literature class in college; does anyone expect me to be right?" But he's worried that my wrongness might reflect poorly on the field of DH, which has a history of critics straw-manning offhand blog posts into terrible representatives of the field. So let me say up front: Persons attempting to find an argument in this post will be prosecuted; persons attempting to find political advocacy in it will be banished; persons expecting me to have anything above a high-schooler's knowledge of English literature will be shot.

Take Huck Finn. In hazy recollection (I haven't read the whole book in probably 10 years), much of what seems great about it is the purely American picaresque of a vision of America. Twain's interest "is in the the boy in whose mouth he puts the story, and in this boy's view of the world as it passes under his eye." Huck "is a true child of the river," and gives us a view of America seen through the eyes of "a perfect vagabond of a youngster, wandering up and down the river at his will, taking in the passing show with open mind, finding it all for to admire."

All those quotes, as you may already have guessed guessed, are not describing Huck Finn at all, but instead come from a review of Charles Stewart's Partners of Providence (1904).


Read the book online through Hathi

The table of contents is pretty fascinatingly close to Huckleberry Finn; the reviewers note the comparison, and it's hard to imagine that the tale of a young boy's adventures up and down the river with an entertaining ethnic (here Irish) sidekick past swindlers and exhibitions and perils wasn't somehow noveled on the most famous humorist in the country.
But there are surely differences as well; I wouldn't be surprised if an in-class discussion on the racial politics Huckleberry Finn couldn't benefit from a brief comparison to Partners' account of "the marooning and subsequent escape of a pair of pugnacious darkies."

Across the c. 4.5 million public domain volumes in the Hathi Trust, there are a surprising number of these, many books that seem (based on Google searches) to languish in deserved obscurity. (I've got a set of tricks that actually finding the pairings more feasible than running 20 trillion pairwise comparisons, but the exact mechanics of that are for another day). But they're interesting; not in a "distant reading" way, but in that they provide some greater focus around the core texts we all read already.


So let me just plug a few books in here and see what comes back. My criteria are just that the original book be canonical.

Huckleberry Finn

Twain is closest to himself; Huck Finn is closest to the later Tom Sawyer books than to Tom Sawyer itself, which should perhaps not be surprising.

But nearest-neighbor searching also reveals a deep vein of western boys literature. We know that this exists; the interesting questions here would probably involve the specific ways (especially dialect: these are mostly first person narratives in highly vernacular styles) that writers imitate Twain.

Publication years also provide a point of departure. All the books here were written substantially later than Huck Finn except for "Live boys in the Black Hills." So if I were going to pick any up, maybe I'd start there.


Moby-Dick

This has fewer straightforward imitators; but the whaling novel is a perfectly well-represented genre.
The closest match is the romance "The Red Eric; or, The whaler's last cruise. A tale" from 1883. Some elements of the contents are provocative, at least; but the similarities are less than perfect. (Red Eric's captain's "insane resolution" is to bring his daughter on a whaling cruise with him, for example).




A few other options include a collection of sea stories,


The Cruise of the Cachalot and Sea-wrack, by Frank Bullen, offer some of the more interesting comparisons. Properly shuffled, it makes sense that Moby Dick's closest companions might include not literature at all, but piecemeal miscellanea from the magazines like this ("Sea-Wrack")


Middlemarch

Middlemarch is somewhat harder to find close matches for uninteresting reasons: since the novel is so long, it was frequently chopped into 2, 3, or 4 parts; and each one of those sections ranks highly on the list.
The nearest novels are by Dinah Craik, who I don't know, but who seems well enough established as a poor man's George Eliot in the scholarly literature. (Googling quickly brought me to the online version of Sally Mitchell's monograph on the author.). "Hannah", the closest, is characterized by Mitchell as "a one-issue novel with a narrow legislative aim."

Fraternity; a romance ... (1910) is a harder nut to crack. It's a rural novel set in Wales and published by Macmillan around 1888, but the only surviving digital copy was (according to library metadata) published in the United States in 1910. (Galsworthy's 1911 novel Fraternity further muddies things here.) It's the subject of a strikingly positive review in the Boston press that explicitly casts it as a diamond in the rough.



I was going to let it go there, but then discovered a whole separate track via this book. The author is one Miss M. M. Holland Thomas, and the novel somehow attracted the intense admiration of JP Morgan for its message of social reform through benevolent patronage. (It is Morgan who paid for the American reprint in 1910.) Does this story have anything to do with a similarity to Middlemarch? Hmm. there's definitely something here about the connections between the English social novel and political intentions. But beyond that, I couldn't say.


The Education of Henry Adams

The absolute closest match is his brother's autobiography. Which should surprise no one, and I'm sure I've encountered the book before. "Early Memories" by Henry Cabot Lodge is also high on the list, which is probably a decent choice as well. But I'll pick as the dopplegänger Cambridge Sketches by Frank Preston Stearns, which hits a number of the same points

The Souls of Black Folk

A real genre-bender of a book, even more than Moby Dick. And even less often reprinted.

The closest match is a fairly dull-seeming hagiography of Booker T. Washington. But I'll take as a shadow "Up stream: an American chronicle" by Ludwig Lewisohn. It seems to be the personal memoir of a German-born Jew who grew up in Charleston, SC before attending Columbia and (eventually) becoming a founding faculty member at Brandeis. The grounds for similarity aren't entirely clear--perhaps some odd combination of self-recognition, music, and the South?--but that's what makes it an interesting track. Some of the 


Autobiography of an ex-colored man

On the topic of great Af-Am literature. This one was suggested to me as a candidate by John Reuland. For this one I'm pasting in a longer list of matches, because we were initially very disappointed at the results. (Very little African American literature on the list).

But on looking at the list, what there is is an extraordinary amount of autobiographical self-help literature about money. So maybe there's some lesson to be gleaned there.


OK, that's enough.

Portrait of the Artist as a Young Man

Again, the matches aren't as clear; a vocabulary-based approach like mine works best thematically distinct themes like riverboats, not with "childhood."

There are some vaguely interesting similarities: at #3, I particularly like "What to read at Winter Entertainments," in which it appears the closest antecedent to Joyce is a stuffed-together hodgepodge of great British writers from the 19th century. Sounds about right.

But as a Doppleganer, I'll take Shaw Desmond's Gods, which seems to cover similar places in the Irish experience of the early 20th century.


On Interestingness

I've thinking about Ted Underwood's "old-fashioned, shamelessly opinionated, 1000-word blog post" from yesterday. There are parts I wholeheartedly agree with, such as the section where he dances near to, but decorously avoids citing, Kieran Healy's magnum opus on what calls for nuance do in contemporary academic discourse. There are parts I don't; I'm increasingly convinced that efforts to apply and invent novel algorithmic practices should be fully central to the work of some humanists, and that calls to return to the primary questions of the disciplines are not just premature but somewhat misguided.*

(Roughly, although I should boil this up into a richer stew at some point: very few people outside a philosophy department think that only academic philosophers should do philosophy; very few people *inside* history departments think that only academic historians should do history. Just as we let political philosophy flourish in politics departments and cultural history flourish in art and music departments, computer programming shouldn't be the sole province of computer science departments.)

Is this interesting? I'm not sure. It's not here-I-come-PMLA interesting, for sure. But then again, I've never deliberately sought out much contemporary literary history written since 1980 or so. For a certain sort of Arnoldian prudish conception of literature, I kind of like the game. Much like my anachronism-searching blog posts, it's a field-and-context approach to literature where the whole is not treated as the object of study itself (the stated purpose of much "distant reading") but as a conveniently large wall on which to reposition the works of literature we're already interested in. What that means for literary history, I think I'm under no professional obligation to say.

Bonus links

A little bonus for those who read through to the end; a temporary link to a live interface to the engine I used for this thing, so you can play along at home. Just go to http://benschmidt.org/similarities/ and you can paste in any text you're interested in. Terms and conditions are: don't link to that page, because this may not scale; and e-mail me or post in the comments if you find any terrible bugs or interesting matches.

Tuesday, November 3, 2015

Word embedding models

A heads-up for those with this blog on their RSS feeds: I've just posted a couple things of potential interest on one of the two other blogs (errm) I'm running on my own site.

One, "Vector Space Models for the digital humanities," describes how a newly improved class of algorithms known as word embedding models work and showcases some of their potential applications for digital humanities researchers.

The other, "Rejecting the gender binary," is a more substantive look at how the method can help us better imagine a version of English without gendered language through some tricks of linear algebra; that results in a sort of translation dictionary between the way students talk about men and the way they talk about women.

I'm aware that this blog is sort of twisting on the vine right now. I like the politics of not using Google, and the ability to embed real javascript that comes with not using Blogger. Perhaps the humane thing to do would be retire this site and direct you to http://benschmidt.org/posts/ and http://bookworm.benschmidt.org instead. But I like keeping it around, and will probably come back here next time I have something to say about, say, the hilariously inadequate college rankings the Economist just published, or just to link other stuff.

Monday, January 19, 2015

State of the Union--and corpus comparison.

Mitch Fraas and I have put together a two-part interactive for the Atlantic using Bookworm as a backend to look at the changing language in the State of Union. Yoni Appelbaum, who just took over this week, spearheaded a great team over there including Chris Barna, Libby Bawcombe, Noah Gordon, Betsy Ebersole, and Jennie Rothenberg Gritz who took some of the Bookworm prototypes and built them into a navigable, attractive overall package. Thanks to everyone.

The first part is an interactive map with every place name we could find using the Stanford Natural Language Toolkit and some (Fraas-flavored) elbow grease. Then we got two great historians of American foreign policy, Dael Norwood and Gretchen Heefner, to explain some of the things in the maps.

The second is about individual words presidents use. So the recent rise in "Freedom," the references to the Constitution predominantly in the time of crisis, and so forth.

My favorite feature, and one that the Atlantic team executed beautifully, is the deep access into individual texts: click on a circle or a bar, and you are off reading the actual paragraph from the state of the union that uses that word on mentions that place. This has always been a core feature of Bookworm on various levels--by treating paragraphs as documents for the modelling, it's easy to drill straight to the interesting stuff. One thing that's mostly missing are the Ngrams-style line charts. I've been saying for a while that I hope people see Bookworm enabling other forms of visualization. These pages are a great example of that; maps and bar charts of words are just as engaging, and sometimes things like "presidents" and "the world" are more engaging than individual years.

So go check those out. They speak for themselves.

But for the text analysis crowd, I also wanted to tell you a little more about the link down right at the bottom of the second (words) piece, and get a little technical about why that, although we decided not to include on the Atlantic site, contains the germ of something I find pretty interesting for online text analysis in general.

Tuesday, December 30, 2014

Federal College Rankings: The pitfalls of a magical regression model

Far and away the most interesting idea of the new government college ratings emerges toward the end of the report. It doesn't quite square the circle of competing constituencies for the rankings I worries about in my last post, but it gets close. Lots of weight is placed on a single magic model that will predict outcomes regardless of all the confounding factors they raise (differing pay by gender, sex, possibly even degree composition). As an inveterate modeler and data hound, I can see the appeal here. The federal government has far better data than US News and World Report, in the guise of the student loan repayment forms; this data will enable all sorts of useful studies on the effects of everything from home-schooling to early-marriage. I don't know that anyone is using it yet for the sort of studies it makes possible (do you?), but it sounds like they're opening the vault just for these college ranking purposes.

The challenges raised to the rankings in the report are formidable. Whether you think they can work depends on how much faith you have in the model. I think it's likely to be dicey for two reasons: it's hard to define "success" based on the data we have, and there are potentially disastrous downsides to the mix of variables that will be used as inputs.

Federal college rankings: who are they for?

Before the holiday, the Department of Education circulated a draft prospectus of the new college rankings they hope to release next year. That afternoon, I wrote a somewhat dyspeptic post on the way that these rankings, like all rankings, will inevitably be gamed. But it's probably better to bury that off and instead point out a couple looming problems with the system we may be working under soon. The first is that the audience for these rankings is unresolved in a very problematic way; the second is that altogether two much weight is placed on a regression model solving every objection that has been raised. Finally, I'll lay out my "constructive" solution for salvaging something out of this, which is that rather than use a three-tiered "excellent" - "adequate" - "needs improvement", everyone would be better served if we switched to a two-tiered "Good"/"Needs Improvement" system. Since this is sort of long, I'll break it up into three posts: the first is below.


Thursday, December 18, 2014

Administrative layers

Sometimes it takes time to make a data visualization, and sometimes they just fall out of the data practically by accident. Probably the most viewed thing I've ever made, of shipping lines as spaghetti strings, is one of the latter. I'm working to build one of the former for my talk at the American Historical Association out of the Newberry Library's remarkable Atlas of Historical County Boundaries. But my second ggplot with the set, which I originally did just to make sure the shapefiles were working, was actually interesting. So I thought I'd post it. Here's the graphic: then the explanation. Click to enlarge.



Tuesday, December 16, 2014

Fundamental plot arcs, seen through multidimensional analysis of thousands of TV and movie scripts

Note: a somewhat more complete and slightly less colloquial, but eminently more citeable, version of this work is in the Proceedings of the 2015 IEEE International Conference on Big Data. Plus, it was only there that I came around to calling the whole endeavor "plot arceology."

It's interesting to look, as I did at my last post, at the plot structure of typical episodes of a TV show as derived through topic models. But while it may help in understanding individual TV shows, the method also shows some promise on a more ambitious goal: understanding the general structural elements that most TV shows and movies draw from. TV and movies scripts are carefully crafted structures: I wrote earlier about how the Simpsons moves away from the school after its first few minutes, for example, and with this larger corpus even individual words frequently show a strong bias towards the front or end of scripts. These crafting shows up in the ways language is distributed through them in time.

So that's what I'm going to do here: make some general observations about the ways that scripts shift thematically. In its own, this stuff is pretty interesting--when I first started analyzing the set, I thought it might an end in itself. But it turns out that by combining those thematic scripts with the topic models, it's possible to do something I find really fascinating, and a little mysterious: you can sketch out, derived from the tens of thousands of hours of dialogue in the corpus, what you could literally call a plot "arc" through multidimensional space.


Words in screen time

First, let's lay the groundwork. Many, many individual words show strong trends towards the beginning or end of scripts. In fact, plotting movies in what I'm calling "screen time" usually has a much more recognizable signature than plotting things in the "historic time" you can explore yourself in the movie bookworm. So what I've done is cut every script there into "twelfths" of a movie or TV show; the charts here show the course of an episode or movie from the first minute at the left to the last one at the right. For example: the phrase "love you" (as in, mostly, "I love you") is most frequent towards the end of movies or TV shows: characters in movies are almost three times more likely to profess their love in the last scene of a movie than in the first.

Thursday, December 11, 2014

Typical TV episodes: visualizing topics in screen time

The most interesting element of the Bookworm browser for movies I wrote about in my last post here is the possibility to delve into the episodic structure of different TV shows by dividing them up by minutes. On my website, I previously wrote about story structures in the Simpsons and a topic model of movies I made using the general-purpose bookworm topic modeling extension. For a description of the corpus or of topic modeling, see those links.

Note: Part II of this series, which goes into quantifying the fundamental shared elements of plot arcs, is now up here.

In this post, I'm going to combine those two projects. What can we see by looking at the different content of TV shows? Are there elements to the ways that TV shows are laid out--common plot structures--that repeat? How thematically different is the end of a show from its beginning? I want to take a first stab at those questions by looking at a couple hundred TV shows and their structure. To do that, I:

1. Divided a corpus of 80,000 movies and TV show episodes into 3 minute chunks, and then divided each show into 12 roughly-equal parts.
2. Generated a 128-topic model where each document is one of those 3-minute chunks, which should help the topics be better geared to what's on screen at any given time.
3. For every TV show, plotted the distribution of the ten most common topics with the y-axis roughly representing percent of dialogue of the show in the topic, and the x-axis corresponding to the twelfth of the show it happened in. So dialogue in minute 55 of a 60-minute show will be in chunk 11.

First a note: these images seem not to display in some browsers. If you want to zoom and can't read the legends, right click and select "view in a new tab."

Let's start by looking at a particularly formulaic show: Law and Order.





The two most common topics in Law & Order are "court case Mr. trial lawyer" and "murder body blood case". Murder is strongest in the first twelfth, when the body is discovered; "court case" doesn't appear in any strength until almost halfway through, after which it grows until it takes up more than half the space by the last twelfth.

That's pretty good straight off: the process accurately captures the central structuring element of the show, which is the handoff from cops to lawyers at the 30 minute mark. (Or really, this suggests, more like the 25 minute mark). Most of the other topics are relatively constant. (It's interesting that the gun topic is constant, actually, but that's another matter). But a few change--we also get a  decrease in the topic "people kid kids talk," capturing some element of the interview process by the cops; a different conversation topic, "talk help take problem," is more associated with the lawyers. Also, the total curve is wider at the end than at the beginning; that's because we're not looking at all the words in Law & Order, just the top ten out of 127 topics. We could infer, preliminarily, that Law and Order is more thematically coherent in the last half hour than the first one: there's a lot of thematic diversity as the detectives roam around New York, but the courtroom half is always the same.

Compare the spinoffs: SVU is almost identical to the Law & Order mothership, but Criminal Intent gets to the courtroom much later and with less intensity.






See below the fold for more. Be warned: I've put a whole bunch of images into this one.

Monday, September 15, 2014

Screen time!

Here's a very fun, and for some purposes, perhaps, a very useful thing: a Bookworm browser that lets you investigate onscreen language in about 87,000 movies and TV shows, encompassing together over 600 million words. (Go follow that link if you want to investigate yourself).

I've been thinking about doing this for years, but some of the interest in my recent Simpsons browser and some leaps and bounds in the Bookworm platform have spurred me to finally lay it out. This comes from a very large collection of closed captions/subtitles from the website opensubtitles.org; thanks very much to them for providing a bulk download.

Just as a set of line charts, this provides a nice window into changing language. I've been interested in the "need to"/"ought to" shift since I wrote about it in Mad Men: it's quite clear in the subtitle corpus, and the ratio is much higher as of 2014 than anything Ngrams can show.

Add caption