Text Analyzer Reveals Emotional 'Temperature' of Novels and Fairy Tales

KentuckyFC writes "Stories are a powerful channel for communicating emotions. But while they have been studied in detail by generations of critics, there is little in the way of objective tools for analyzing and comparing their emotional content. That looks set to change thanks to one data mining researcher who has applied the process of sentiment analysis to novels and fairy tales that have been digitized on Project Gutenburg and the Google Books Corpus. The results show the density of emotions in different parts of a story and how the emotional 'temperature' changes throughout the tale. For example, this guy has used the technique to compare the emotional content of the entire collection of the Brothers Grimm fairy tales to reveal that the darkest story is a tale called Gambling Hansel; clearly a lesson to us all."

Comments

I usually start with a Gambling Hansel....

[Anonymous Coward]

then I move to a Dirty Gretel and finish all off with a move I call the Sugar Fairy Plums.

Ironic captcha: bodice

Re:I usually start with a Gambling Hansel....

[alphatel]

Sadly Gambling Hansel isn't nearly as dark as Hansel and Gretel or Hans in Luck. So I guess the algorithm sucks?

This Guy

[sexconker]

The summary doesn't even mention the researcher's name? I mean, I agree that this is useless, pointless "research". But if you're going to piss about and drop Project Gutenburg and "Google Books Corpus" which are only tangentially related, couldn't you at least give "this guy" a fucking name?

Re:This Guy

[mythosaz]

Ahhhh.... This guy... *points at guy...*

Re:This Guy

[mugurel]

It's Project Gutenb*e*rg

Re:This Guy

[sexconker]

It's Project Gutenb*e*rg

And that's what I initially typed. I had to force myself to copy Gutenburg from the summary.

It's a Pot Boiler!

[ackthpt]

"How yer know that?"

"It's written in charcoal."

i'll get me coat

The "eight fundamental emotions"

[PapayaSF]

From TFA:

Analysing the emotional content of text is also becoming easier. In recent years, researchers have built up significant databases of the emotions that a given word evokes. This is part of the new field of sentiment analysis in which common words are categorised as positive, negative or neutral and associated with one of the eight fundamental emotions—joy, sadness, anger, fear, trust, disgust, surprise and anticipation.

I don't know about anyone else, but I found that bit as fascinating as the text analyzer itself. But where does laughter fit? Shouldn't it count as a fundamental emotion? Or is it considered just a sub-category of "surprise" or "joy"?

In any case, I wonder if someone could combine all that with the 36 dramatic situations and a few other components, and create a program that writes stories....

Re:The "eight fundamental emotions"

[AthanasiusKircher]

I don't know about anyone else, but I found that bit as fascinating as the text analyzer itself. But where does laughter fit?

Yeah, while this theory of emotions surely has some good aspects, forcing ALL emotionally charged words into these categories will obviously skew the data in certain ways. When a model like this is used to classify something much more complex, the ultimate data analysis often tells you more about the model than about the data. Are we actually tracking the changes in "sadness" and "fear" over the course of Hamlet, or are we tracking some arbitrary dividing line that this model forces us to use to classify words?

Moreover, these sorts of "digital humanities" projects that just analyze a corpus by counting up occurrences of words are always incredibly limited. It's so easy to skew the data just because a character or place or something in the story happens to include a word that is "emotionally charged" in a particular way in this model.

For example, I note that "Godfather Death" is nearest to "Gambling Hansel" in terms of the "darkest story" in the study. While the story is dark, it's actually rather short and a very simple moral, with a main character who is a remarkable healer. The main character doesn't do too well at the end, but in the gamut of Brothers Grimm stories, this one doesn't have a lot of "dark" details. I assume, instead, that this story gets rated as very "dark" because one of the main characters happens to be named "Death," a word that has a very negative valence in the model. Write a happy story at a bar called "The Good Death" (referencing either bravery or sexuality), include a few other character names or place names that recur frequently but just happen to sound "negative," and I bet this algorithm will judge it "dark" too... or at least not as "positive" as the plot would suggest.

I'm not saying such studies are useless. But they really need to factor in context, multiple meanings, and especially other factors that might lead to high frequencies of their chosen "emotional" words, like proper names or other plot points that may not actually be representative of the vocabulary and emotions of the story overall. In essence, for anything meaningful to come out of word frequency studies, you actually need to read the text as well and take account things that would obviously skew the data.

Re:The "eight fundamental emotions"

[PapayaSF]

But they really need to factor in context, multiple meanings, and especially other factors that might lead to high frequencies of their chosen "emotional" words, like proper names or other plot points that may not actually be representative of the vocabulary and emotions of the story overall.

You are correct. Obviously this sort of text analyzer is still in its infancy. It would be interesting to throw some oddball stories at it and see the results. E.g., here's a story filled with unpaired words. I wonder what it would say its "emotional temperature" was? And of course the program would totally miss the humor. (Note that the New Yorker blew the formatting when they put this online, and that the actual story starts with the third sentence: "It had been a rough day....")

Re:The "eight fundamental emotions"

[sconeu]

That's because the authors couldn't get no satisfaction, though they tried and they tried, and they tried, and they tried.

Re:The "eight fundamental emotions"

[phantomfive]

I think laughter is a response to an emotion, not an emotion itself. Thus you can have nervous laughter, or joyful laughter, or laughter of relief.........

Re:The "eight fundamental emotions"

[PapayaSF]

Hmmm, interesting. But what is laughing at a joke? It isn't necessarily nervous, or joyful. It's sort of a relief, in that the tension of the buildup is released in the punch line, but that doesn't seem to be the same as "laughter of relief." Humor, or laughter, just seems to me to be as core of an emotion as fear or anger, but maybe psychologists and data miners don't see it that way.

Re:The "eight fundamental emotions"

[dougisfunny]

That is a rather limited way to look at satisfaction. Relief, catharsis, vengeance, schadenfreude, comeuppance to name a few could all be descriptive of something satisfactory, but might not cross over into joy so much.

Re:The "eight fundamental emotions"

[hyperfine transition]

From TFA:

In any case, I wonder if someone could combine all that with the 36 dramatic situations and a few other components, and create a program that writes stories....

Someone has ... every Hollywood studio has it running on their server farms

Re:The "eight fundamental emotions"

[Panoptes]

"This is part of the new field of sentiment analysis in which common words are categorised as positive, negative or neutral and associated with one of the eight fundamental emotions - joy, sadness, anger, fear, trust, disgust, surprise and anticipation."

Codswallop! This notion has been around for a very long time under the name of connotation. Giving something a new name and peddling it as a new concept doesn't exactly inspire confidence in the writer's competence or integrity.

Think of the Children

[Froeschle]

After reading this article I can only say that I am grateful to have my own childhood behind me.
[the end]

Tropes

[PRMan]

Let me know when it can call out all the TVTropes in a story...or would that cause an endless loop?

Re:Tropes

[vlueboy]

Let me know when it can call out all the TVTropes in a story...or would that cause an endless loop?

This troper believes that the fun in visiting tvtropes is knowing other geeks have enjoyed each story... or *hated* it, and in many other ways been driven to break down the story for your enjoyment due to resonating with that story emotionally. Call it pride in exclusivity, sort of like coming to slashdot looking for fellow geek observations, tips, jokes, and so on. In short, tvtropes is fun because you're finding someone else that you share some knowledge with in a way the corresponding Wikipedia article cannot fulfill.
So... once a computer is doing the "thinking", even making a knowledge tree 1000 times denser than the depth of TVtropes, we'd know there is no emotion in it, and that it won't be growing on its own from outside contributions, or real-life anecdotes tangentially related to the content, or even grow (would the computer be "watching" new series and recalculate everthing, or would it be too busy maintaining a finite set of data frozen in time.) It COULD be done given enough tech, but knowing implementations, it'd feel like landing at some endless Google linkbait farm that links endlessly to itself, and TRYING to force yourself to enjoy the crickets chirping while you click.

Re:Tropes

[Zanadou]

This troper...

There's a trope for that.

Here is the PowerPoint for the paper

[TedTschopp]

Here is the a good summary of the work in a PDF of a PPT.

http://www.saifmohammad.com/WebDocs/LaTeCH-emotions-in-books.pdf

Ted

Re:Here is the PowerPoint for the paper

[retchdog]

1. used Mechanical Turk to get people to report association of words with emotions
2. determine emotion by counting(!) corresponding words with weighting proportional to association, using sliding window in time.
3. generate pretty but almost meaningless plot
4. profit?

hint: Dramatic tension is often created through irony. The audience knows that the doom of character X (e.g. Walter White) comes when he finally comes to trust character Y (e.g. the white nationalist thugs) implicitly. Goddammit, at the very least look for some negative auto- or cross-correlations.

Re:Here is the PowerPoint for the paper

[tinkerton]

Here's a good basic emotion measure: adjective count. It covers the bulk of written text even though one can find ways around it.

Gambling Hansel: not dark at all

[themushroom]

I didn't find it dark at all, not nearly as dark as the tales Disney sanitized. I mean, it's about a gambler who beats both God and the Devil even if he has lousy luck with mortals prior to getting rigged cards and dice.

Re:Gambling Hansel: not dark at all

[gl4ss]

well it's pretty dark that seemingly people on earth have no choice but to gamble nor do beings in hell and heaven.

other than that, it's pretty light hearted for a grimm.

there's even 7 years in gambling hansel during which nobody dies, since he beat death too - and his soul lives eternally in gamblers.

so the algorithm is pretty contextually unaware. maybe it just counts nots and buts. waste of time anyhow, even if it did get me to read gambling hansel.

Re:The Bible?

[tepples]

Wouldn't the steamiest book of the Bible be Fifty Shades, umm, Song of Solomon?

Related stuff by my wife on tagging narratives

[Paul Fernhout]

Mainly by hand though. Free book: http://www.workingwithstories.org/
Free software for communtieis: http://www.rakontu.org/
Related business process patent (sadly) when at IBM Research: http://www.google.com/patents?hl=en&lr=&vid=USPAT7136791
Past commercial software: http://www.sensemaker-suite.com/
National security (does have some automatic aspects): http://app.rahs.gov.sg/public/www/content.aspx?sid=2955

There is a lot you can do with stories once they are tagged for emotional intensity, whether automatically, by the teller, or by other people. Stories are all around us, as we try to make sense of our lives and events in our communities. So this sort of technology to tag emotions in stories is much more far reaching than just being about fiction. It can be used to design better products, to help communities figure out what to do about a pressing issue, to resolve conflicts, and to see emerging trends. That is one reason such work is funded by the intelligence sector (as well as businesses and some non-profits). She's been trying to make these ideas freely available to everyone, but it has been a slow going slog to follow the path of free and open source for all this.

By someone else on the relation between emotion and reason:
http://en.wikipedia.org/wiki/Descartes'_Error

Re:The Bible?

[retchdog]

my favorite book of the Bible is Ecclesiastes which, being basically an existential musing on the meaninglessness of life by, ostensibly, King Solomon, is considered so out-of-place that scholars have been trying for about two thousand years to figure out why the hell it was included in the Hebrew canon.

Re:The Bible?

[TedTschopp]

The best answer that I have heard is that the existential nihilism that is covered by the book is an important aspect of Jewish / Christian traditions and that all wise people must confront it and think about it. The idea is so central that it even suggests the idea that God himself wrestles with this question and more specifically in the Christian Tradition this is what the Christ wrestled with on the cross when he cried out "Why have you forsaken me?" The saints, holy people, and mad men through out history have all struggled with this and were all changed by the questions asked by this book.

Re:The Bible?

[__aaltlg1547]

You misunderstood the comment. TedSchopp is saying these are such fundamental questions that every Jew and Christian grapples with, even the Christ who is regarded by Christians as the incarnation of God. The saying of Jesus on the cross "Why have you forsaken me?" is very much of a piece with the existential despair of Eclesiastes. What could be worse for Jesus at that point than feeling that not only is he dying in this horrific way, but that it is a completely meaningless experience, as is the whole of his life and maybe nobody even cares what is happening to him?

Christianity is supposed to be an answer to the questions raised by Ecclesiastes. It is all about life not being futile in the big picture and God actually caring what happens to people. Without the thoughs and feelings expressed in Ecclesiastes, one would have to ask what Christian salvation is even for.

I find it the most universal of all the Judeo-Christian scriptures. It's not just Jews and Christians who must grapple with the apparent meaninglessness of life and the futility of our actions and desires. Everyone must. Every religion and philosophy, if it's worth anything at all, must address these issues.

Text Analysis...

[David_Hart]

So, this analyses text and emotional connotation of words to produce an emotional score for each story. Yet, it has no way of divining context, whether or not a particular section of the story is funny, or if a death causes an emotional reaction of sadness or satisfaction (i.e. the character was evil, deserved it, etc.). In other words, it's an arbitrary system that may work at a basic level but will still get a lot of things completely wrong... at least it's a start, I guess...

Simple

[wonkey_monkey]

if (text contains 'Natalie Portman' && text contains 'grits') temperature='steamy';