Tutorial on Creative Twitterbots

RobotComix.com
For more comix on creativity:

We are going to explore the mechanics of a verbal acrobat: a creative, metaphor- generating Twitterbot. But first we need to ask what makes language creative. We need to ask a critic who pulls no punches: step up Mr. George Orwell …
I’ll take
it from here.

Most people who bother
with the matter at all would admit that the English language is in a bad way.
It becomes ugly and inaccurate because our thoughts are foolish, but the slovenliness of our language makes it easier for us to
have foolish thoughts.

A scrupulous writer, in every sentence that he writes,
will ask himself at least four questions, thus:
What am I trying to say? What words will express it? What image or idiom will make it clearer? Is this image fresh enough to have an effect?

They will construct your sentences for you–even think
your thoughts
for you.
But you are not obliged to go to all this trouble.
You can shirk it by simply throwing your mind open and letting the ready-made phrases come crowding in.

This invasion of one's mind by ready-made phrases can only be prevented if one is constantly on guard against them, and every such phrase anesthetizes a portion of one's brain.

Marcel Duchamp’s “Fountain” (1917): The Art World’s Most Famous Ready-Made Artwork
Sigh. Prose consists less and less of WORDS chosen for their meaning, and more and more of PHRASES tacked together like the sections of a pre- fabricated hen-house.

There is a huge dump* of worn-out metaphors which have lost all evocative power and are merely used because they save people the trouble of inventing phrases for themselves.
Moreover ….
*Though linguist Geoff Pullum calls Orwell’s polemic “stupid and dishonest”, a “load of Orwellian cobblers”

There is a long list
of fly-blown metaphors which could be got rid of if enough people would interest themselves in the job. Sigh

This [filtering] effort of the mind cuts out all stale or mixed images, all pre-fabricated phrases, needless repetitions, and humbug and vagueness generally.
It is better to put off using words as long as possible and get one's meaning as clear as one can through pictures or sensations.

When one watches
some tired hack
mechanically repeating the familiar phrases, one often has a curious feeling that one is not watching a live human being but some kind of dummy.
Hey!

A speaker who uses that kind of phraseology has gone some distance towards turning himself into a machine.
The appropriate noises are coming out of his larynx, but his brain is not involved as it would be if he were choosing his words for himself.
There is a new sheriff in town …

The worst thing
one can do
with words is to surrender them.
What is above
all needed is to
let the meaning choose the word, and not the other
way about.

Never use a metaphor, simile or other figure of speech which you are used to seeing in print.
Not even in cut-ups?
Cutting up print norms is a wonderful basis for linguistic creativity!

If it is possible
to cut a word out, always cut it out!
Never use a
foreign phrase, a scientific word or a jargon word if you can think of an everyday English equivalent.
Keep It Simple, Stupid …

If it is possible
to cut a word out, always cut it out!
So when you make a stupid remark its stupidity will be obvious, even to yourself.

Costly thy habit as thy purse can buy,
But not expressed in fancy
—rich, not gaudy,
For the apparel oft proclaims the man.
.
Readers tend to find a text more profound, and dig deeper to give it a meaning IF …
… it has
a clearly poetic form. Psychologists call this The Keats Heuristic.

The Keats Heuristic at work:
“If the glove don’t fit you must acquit”
Short, catchy texts can have a profound effect on how we perceive the “truth”.

A scene from the film Mystery Men: Our ragtag “super”-heroes are preparing their homemade costumes for the big showdown when The Sphinx’s over-use of Chiasmus begins to grate on his team-mate Mr. Furious …
The Sphinx: He who questions training, only trains himself in asking questions. […] Work well on your new costumes my friends, for when you care for what is outside, what is inside cares for you. […] To summon your power for the conflict to come, you must first have power over that which conflicts you.
Mr. Furious: Okay, am I the only one who finds these sayings just a little bit formulaic? "If you want to push something down, you have to pull it up. If you want to go left, you have to go right." It's …
The Sphinx: Your temper is very quick, my friend.
But until you learn to master your rage …
Mr. Furious: … your rage will become your master?
That's what you were going to say. Right? Right?
The Sphinx: um … Not necessarily.

A creative bot must not rely then on a single rhetorical strategy: it should be able to draw on a diversity of forms and strategies. In other words:

A (Web-) Service-Oriented Architecture is “an architectural model that aims to enhance the efficiency, agility, and productivity of an enterprise by positioning services as the primary means through which solution logic is represented”
Erl (2008)
New Metaphor Services should be discoverable, autonomous and widely reusable, and should be flexible enough to compose in groups, while remaining loosely coupled to others. Services should also maintain minimal state information and use abstraction to hide the complexity of their inner workings and data.

So what if a diverse assortment of metaphors could be created on demand for any given topic?
A Web Service called Metaphor Magnet can be used to commoditize creative metaphors for use on Twitter via bots.
Check out: Afflatus.UCD.ie for more details of this and other services.

This bot by Tony Veale uses his Metaphor Magnet Web-service to tweet a new hard-boiled metaphor every hour

Genius?
Guano?
OR
Not every automated metaphor / tweet will hit its target. Our bot will often shoot blanks. So how do we ensure that the bot is mostly generating metaphors that it itself understands and appreciates?

The Twitterbot Philosophy of “Mere Generation”
Even a stopped clock is right twice a day.
So even a random Twitterbot will eventually tweet something that seems sensible and perhaps even profound. (The Keats Heuristic helps.)

The world will often provide a chance context that makes a random message seem timely and apt. Consider this tweet from Adam Parrish’s @everyword bot, which just tweets single words!
Why was this one-word tweet worth re-tweeting 243 times? Because it coincided with the sacking of the NY-Times first female chief editor. Spooky huh!

Humans are
constantly looking
for order, meaning
and relevance. A bot’s audience may perceive aptness even where
it is not intended.
All things being equal, a twitterbot should earn this perception of aptness and insight. But we humans are ready to grant it anyway, even if only “playing along”.

Chance occurrences are not viewed as simple coincidences, but as acts of fateful serendipity, by an audience eager to find meaning and relevance.
You’ll have to take my word for this – though do check the date of the tweet – but the above output was generated as this tutorial was being written. OMG #not

In the long run though … while a mindless bot may produce some rare gems amongst its torrent of sheer randomness, how many users would care to be twitter-bombed in this way?

Steering a Creative Course between Pastiche and Mere Generation

“Mere” Generation: Generating Everything is EASY (as in the Borges story “The Library of Babel”). Selective Generation is HARD.

Most twitterbots fall into the category of Mere Generation. They are typically designed to playfully explore the possibilities of generation under minimalist conditions rather than to employ baroque knowledge structures or to be creatively self-critiquing.
A good example of a minimalist bot is Adam Parrish’s @everyword. This bot is working its way through all the words of English in alpha- betical order, generating one every half hour.
Why was this word re-tweeted 2,342 times? Well, you know why, but hold that thought, we’ll come back to it later …

@everycolorbot is another minimalist and very popular exponent of Mere Generation. Simply, the bot generates a random color hex-code (six hex digits denoting an RGB color code) and a picture of the corresponding color.
Just as Borges’ library of Babel provided every possible book, this bot will eventually tweet every RGB color.

In contrast, Pastiche is “Riskless” Generation that veers too close to its inspiring exemplars, to produce safe variations on a successful theme.

@museumbot by Darius Kazemi tweets a hi-res image and text caption from the catalogue of the Metropolitan Museum of Art, 4 times a day. Think of it as an artistic metronome.
The bot does not generate new images or new captions of its own, but uses existing images/captions to good effect.

… they practically use NO knowledge at all, beyond basic word & pattern knowledge.
To steer a creative middle course, one needs knowledge. But most bots are so lightweight …

@metaphorminute by Darius Kazemi tweets a metaphor-shaped text every two minutes (due to Twitter API limits). This knowledge- lite but fun bot makes maximal use of the Keats Heuristic, to package its often bizarre juxtapositions in the familiar forms of poetic metaphor.
This metaphor bot makes use of Wordnik.com, a language web-site & dictionary with a Web API. Many words provided by Wordnik are rare or abstruse, lending tweets an air of mystery & ersatz profundity.

@twoheadlines by Darius Kazemi uses the Cut-Up Technique developed by William Burroughs & Brion Gysin. Two texts are spliced to promote a jarring bisociation of ideas in the sense of Arthur
Koestler.

Princeton WordNet
Pros: large dictionary/thesaurus; free use; hierarchical ISA structure
Cons: No explicit semantics; uneven coverage; no pragmatics
ConceptNet
Pros: crowd-sourced KB; explicit semantic forms; common-sense
Cons: error-prone; lumpy; uneven coverage; generally unreliable
WWW / Web Corpora
Pros: vast source of implicit beliefs; language in the wild; topical/dynamic
Cons: noisy; unstructured; knowledge must be extracted
Knowledge? What knowledge?

WordNet provides a taxonomic super-structure for its senses, allowing good-old-fashioned AI-style ISA reasoning. It also provides text glosses for each sense, which might be parsed.

ConceptNet is a strange beast: more than a dictionary, less than a formal ontology. Its inventory of semantic forms is often used to encode typical associations, but some very bizarre ones too.

@appreciationbot by Mike Cook demonstrates the potential of bots to exhibit social interaction. This bot responds to tweets by @museumbot with critical appraisals of the artifacts it describes.
This bot uses the MetaphorMagnet CC Web-service to source metaphors on demand (e.g. cylinder as curved blade) and ConceptNet to suggest possible relationships between things. And it uses the Keats Heuristic to evoke a critical discourse.

Web
Linguistic constructs like the simile frame
“as ADJ as a NOUN”
allow us to harvest a rich knowledge-base of stereotypical norms from the Web.  These are from the Google n-grams.

Frequent Web
queries are another
source of a culture’s implicit knowledge. We cannot access them directory, but we can milk them from completions.

Questions are the best source of shared presuppositions that others are assumed to believe also. So we use a trie approach to harvest as many as we can from Google.
We can then extract Subject-Relation-Object triples from these questions for our KB.

‘The book fascinated him, or more exactly it reassured him. In a sense it told him nothing that was new, but that was part of the attraction. It said what he would have said, if it had been possible for him to set his scattered thoughts in order. It was the product of a mind similar to his own, but enormously more powerful, more systematic, less fear- ridden.”
-- That’s from “1984”
But
how can
a bot use all this knowledge to write well, like me?

Orwell puts his finger on a dilemma: “common-sense” KBs aim for uncontroversial and yawn-inducing facts of the world. How can we ever use the contents of these KBs to generate reassuringly meaningful yet fascinating metaphors?
A - I
J - S
T - Z

Consider what we all know about donuts and bagels:
Each is typically round, with a hole in the middle; each is made from flour/dough; each fits in your hand and is tasty to eat. But donuts are typically sweet while bagels are typically savory.
So each is comparable to the other with obvious similarities.
Each also contrasts in significant ways with the other. Thus, one would seem to be the ideal comparison for the other. So how interesting is the obvious metaphor? Very boring indeed!

Unless … we focus on the human-scale qualities of objects.
For instance, donuts are “fun” foods while Bagels are more “serious”.
What about this tweet: A Bagel is just a donut with an office job
“Lazy slacker”
“Busy professional”
So Tip #1: emphasize the human dimension

Tip #2: We have always created “Gods” to personify the concerns that are key to our lives. So use the pattern “God of [X]” to mine a list of the most significant “human-scale” concepts to use in our new metaphors.

Look for all the Google n-grams that match our pattern
“God of [X]”
Web

We find other
building blocks of human culture by using the pattern
“culture of [X]”
Web

I’m Salt
I’m Pepper
salt
and
pepper
724197
Tip #3: use the Google n-grams to find meaningful coordinations of these grand
themes.
Intuitively, we coordinate ideas that belong together. As the linguist Eric Firth famously said,
“You shall a word by the company it keeps.”
Our bot will know ideas by the company they keep.

So notice how
our conceptual “Lego blocks” click together in very interesting ways on the Web.
Web

We can use these coordinations to suggest corpus-guided projections of norms from one topic to another. This is the Metaphor-Eyes service.
Afflatus.UCD.ie

Tip #4:
observe how our Lego blocks click together in existing metaphors in Web n-grams.
We can
build on these.
Yes we can!
Web

competition
is
journey
95
a
Consider two tweets of @MetaphorMagnet. This Web 4-gram competition is a journey
is enriched with some KB triples: champions win competitions; travelers undertake journeys; Champions are energetic; while travelers are usually
exhausted.
Note the deliberate semantic contrast between energetic and exhausted.

Tip #5:
Deliberately combine Lego blocks that represent ideas that are in conflict, and not obviously connected in our
experience.
Case in point:
When I, fashion guru Karl Lagerfeld, compared my perpetual Sunglasses to a Burqa. Oh, how the media loved that one!

Aristotle (in his Poetics) saw metaphor as a question of categories and taxonomies: a metaphor applied the name of one category of things to another
Computer scientists have found this a very attractive view, to apply to IS-A hierarchies and ontologies (e.g. Wilks 1978; Way 1991; Fass 1991; Veale 2003)

How can a static system like WordNet or ConceptNet react like this?
Dark things worn to protect my privacy
The key, of course, is being able to identify, create and name this new super-category.

@MetaphorMagnet uses any opportunity it can to show that two concepts with opposing properties may still reside in the same ad-hoc category. The Google n-grams are a rich source of Web-attested opportunities.
Group terms like “army”, “band” or “gang”, or mass nouns in partitive constructions can put contrasting ideas into the same ad-hoc category. These  pairings are all derived from attested n-grams.

Input Genome #1
Input Genome #2
Blended Genome
Metaphor highlights the shared qualities of two ideas. So a good metaphor can surprise us by gathering and selectively framing what we already know!
Think of our knowledge of a topic as its genetic material. A conceptual blend* splices the DNA of two concepts into a single hybrid
genome.
*Conceptual Blending Theory: Mark Turner & Gilles Fauconnier

A nice example of DNA splicing and Conceptual Blending is offered by the film The Fly, in which a scientist accidentally blends his own DNA with that of a fly.
In metaphor generation we intentionally create just this kind of hybrid.
Though we can often surprise ourselves with the qualities that emerge!

Input #1: POET
Input #2: COMEDIAN
Blend: RAPPER
To craft an insightful metaphor from what we already know, deconstruct a known concept as though it were a deliberate blend of two others concepts.
E.g., observing similarities between Rapper & Poet and between Rapper & Comedian, imagine Rapper as a deliberate blend of Poet & Comedian.
Rapper:
from Poet? Metaphor- composing.
from Comedian? Publicity- garnering.

@MetaphorMagnet uses the Keats Heuristic by packaging the metaphor as a dictionary entry in the style (a bot can dream!) of Ambrose Bierce’s The Devil’s Dictionary.
@MetaphorMagnet
uses the Google n-grams to characterize the blend itself, by finding an attested combo of a target property with a word denoting “mix” or “blending”

To recap: the opposite of what is creative is the obvious, not the familiar. Creativity, in metaphor and elsewhere, extracts new value from old ideas, by pushing us to see what we already know but do not always appreciate.
Truly creative bots must unleash the shock value in the familiar, via the judicious use of novel combinations and pushy inferences, to find where all the bodies are buried!

Causal Blending allows @MetaphorMagnet to interleave the norms of two concepts with differing affect profiles to achieve a surprising whole. Consider Artists: (I used to be an artist!)
By chaining KB triples together, a bot can observe that artists typically cause a positive effect on the arts. Seeing that critics can have a negative effect on arts, and dictators have a negative effect on their critics …
… @MetaphorMagnet reasons that dictators can have the same positive effect on the arts as artists! This is sophistry as ‘critic’ is used in two differing senses, but blend logic wins out: Artists may act like Dictators!

You can’t handle the truth!
So @MetaphorMagnet achieves shock value via causal equivalence: it uses simple reasoning over its knowledge-base to show that two very different concepts lead to the same logical ends.
Sophistry?

“Abstraction may not resonate with many of
@everyword’s followers, who are just hanging around to retweet “sex,” “weed” and “vagina” — at this point, the account’s most-re-tweeted words. But the idea of
using Twitter as a medium for serious art and social commentary has increasingly caught on with a ragtag group of conceptual writers, generative poets, and performance artists.”
-- Caitlin Dewey, The Washington Post, May 23 2014
Why should Creativity researchers build twitterbots?
As noted in the Washington Post, Twitter is emerging as a medium for conceptual art in which users actively seek out autonomous generation. Computational Creativity research can make a ground-breaking contribution.

A good place to start is by creating next-gen CC versions of existing bots that go beyond mere generation.
Consider @everycolorbot – Mike Cook suggests a CC bot that would creatively name its own colors, in a witty parody of Dulux paint names.
Dulux uses pretentious names with positive effect, but a bot might call this one “mushrooms on toast” or “rusty battleship”. Such a bot would exhibit humor and visual appreciation.
This twitterbot could respond to @everycolorbot with novel names, in a game of dueling bots.

Software is NOT human. Software is NOT human. Software is NOT
hum—
Yet it has the power to shock like a human, and even to offend like a human! How can we curtail this power?
Should we even try?

Can a Twitterbot be used as a tool of protest?
Zach Whalen thinks so. His @ClearCongress bot re-tweets the members of U.S. Congress, randomly redacting texts to protest their lack of transparency.
Mark Sample’s protest bot @NSA_PRISMbot mocks the creepy collection of useless “information” by the NSA’s all-seeing PRISM tool as revealed by Edward Snowden.

Sample’s @NRA_Tally bot is more creative, in a CC sense, though it still relies on mere generation.
It generates fake news reports of fictional killing sprees, focusing on the specifics of the firearms used.
“ … not only can protest bots present society a bill it cannot pay, they can do so at the rate of once every two minutes. They do not bullshit around.”
-- Mark Sample, protest-bot creator
@NRA_Tally combines real statistics and place names that are drawn from an FBI database on past spree killings. To emphasize its call for gun control, it also includes a standard NRA response to such events.

So what are you waiting for?
Twitter is a perfect medium for Computational Creativity. A wave of experimental bot builders has already blazed a path with fun, minimalist bots that have attracted users and whetted their appetite for more. Just imagine what CC research can bring to the party!
So C’mon  SkyNet is not going to build itself!

That’ll do, bot -- that’ll do!

Tutorial on Creative Twitterbots

More Related Content

Viewers also liked (16)

Similar to Tutorial on Creative Twitterbots (20)

More from Tony Veale (14)

Recently uploaded (20)

Tutorial on Creative Twitterbots