Project Mashups (predictive text)

Assignment: Use predictive models to generate text: either a Markov chain or an RNN, or both. How does your choice of source text affect the output? Try combining predictive text with other methods we’ve used for analyzing and generating text: use RNN-generated text to fill Tracery templates, or train a Markov model on the output of parsing parts of speech from a text, or some other combination. What works and what doesn’t? How does RNN-generated text “feel” different from Markov-generated text? How does the length of the n-gram and the unit of the n-gram affect the quality of the output?

Original text reference:

Code: to be uploaded

'Intervention to its sequin surfactant towards people in these nuances one places once and 16 vibration be',
'Intervention (active installation to testing and music. Track the fundamentals of white light. The audio ',

 'Interactions and relation from Google Location of collaborative materials; clear. This project uses one p',

 'Interventions: Pet, energetically in reality by the the bigger picture of our minds try to engage with a ',

 'Interactive project between scents of light is asked together). The abstract morphing and plays with dist',

 'Intervention to them. By tuning to testing an associated using the installations creative installation vi',

 'Interactive means towards these are created using p5.js for determining to their motion, time and senses ',

 'Interacts with them. However, we never real time. The installation visuals Process and or uses taken a pr',

 'Interventions; how scents an arduino. Code: Senses agree, our minds ge',

 'Interactive community Day, New York 2019 ITP Wintervention to test human reaction patterns and sense of f']

Textural Lyrics Generator

Assignment: Devise a new poetic form and write a computer program that generates texts that conform to the poetic form you devised.

I began this week’s assignment thinking about what body of text is ‘structurally interesting’. I landed on song lyrics because they are so diverse. When listening to a song, we are not fully aware of its lyrics structure because so much of it is dependent on its delivery i.e. melody, rhythm etc. Looking at the lyrics structure itself, it is almost random with rhymes and emphasis here and there.

I want to experiment and see if I can create a poetic structure that is able to deliver a feeling of a song – something which has a textural feel to it. Something which feels interwoven and ‘lyrical’.

As a result, I created ‘Textural Lyrics Generator’.

The generator (currently me curating the input) takes an album, looks up its song lyrics and breaks them up into lines. It then finds the rhymes at the end of each line and group those which are the same.

Once the lines are grouped (currently running in code), it randoms a title of the song, randoms a number of verses between 3-6 verses, randoms a number of lines between 1-8 lines and ends with credits to the songwriter.

After the process is done, it prints out the lyrics.

After deciding on the structure of the lyrics I started looking for a source text. I took Bon Iver’s ‘For Emma, Forever Ago’ album because I could never hear the words of the songs. When listening to the songs, the words are so in-tune with the melodies that the blend with each other and the voice becomes merely an envelope of sound. This made me wonder if I could create the same feeling with the ‘Textural Lyrics Generator’.

These were the results:


I move in water, shore to shore;
tail on
I move in water, shore to shore;
nothing’s more.
take all on the wind on

cause blinded I was blindsided
the end of a blood line... the moon is a cold light
him: “for every life...”

leaving rope burns— her:

Original text by Justin Vernon



peek in... into the peer in

so ready for us,
I am my mother’s only one

her: “forgo the parable.”
her: “forgo the parable.”

Original text by Justin Vernon



him: “for every life...”

for the agony, I’d rather know
running home, running home,
I’m crippled and slow
I cup the window
now you know.

the so many territories
for the agony
for the agony

teased by your blouse
teased by your blouse
spit out by your mouth
would you really rush out for me now?
spit out by your mouth

Original text by Justin Vernon

so many foreign worlds
“I toured the light; so many foreign roads for Emma, forever ago.”
running home, running home,
contrasting the snow
there’s a pull to the flow

I was full by your count
bike down... down to the downtown
spit out by your mouth

the creature fear

Original text by Justin Vernon


Future development: automate the input curation and sorting process mentioned above.

Blog prompts:

How well does the output of your computer program conform to your invented poetic form? Could a human do it better?

I was very happy with the results because they turned out to be unique yet very meaningful. I believe a human could do it but it would not contain the sense of ‘random’ (which technically is an element related to creativity) and would take a much longer time.

How does your choice of source text (your “raw material”) affect the character and quality of the poems that your program generates?

As the source of text I chose was very poetic to begin with, the results had the quality of a ‘poem’. However, I can imagine this working or more pop lyrics as well, for instance, taking a lady gaga album, or an R&B/soul album such as Steve Wonder’s ‘Songs in the Key of Life’.



Work-in-process results I found interesting:

Screen Shot 2562-03-13 at 23.31.41.png


Assignment: The digital cut-up. Create a notebook program that reads in two or more texts and stores portions of them in Python data structures. The program should create textual output that creatively rearranges the contents of the text. Use functions from the random module as appropriate. You must use lists as part of your procedure. Choose one text that you created with your program to present in class.

My theme for this semester has been to explore the five senses: seeing, hearing, tasting, touching and smelling. Although we perceive our surroundings in multi-dimentions, the studies of these senses are often done separately. This made me wonder how the studies might read if they were combined.

With this initial thought, I set out to find artists specialised in each field and traced their interviews which they spoke about their passionate subjects.

Light – James Turrell
Sound – R. Luke DuBois
Taste – René Redzepi
Smell – Sissel Tolaas

I could not pick an artist who specialised in Touch as I wasn’t aware of any.

I took the interviews from various sources including Cycling 74, Interview Magazine, Designboom and Art Net, and shuffled and rearrange the format of the presentation as shown below.

An interview with James Turrell on Scent by Cycling 74

CYCLING 74: History. How'd you explain here?

JAMES TURRELL: I did my undergraduate degree at columbia. My first-year here i dated
Smell, that which we’ve all as i pay light start better, class a to smell and
sounds, for i at on in so best your can it, was that gas, you were else it i
same that’s i very to i computer, the of actively techniques the a smell took an
assignment in unpredictable it’s and that unexpected the made wh

CYCLING 74: Is the her for your culinary with reason a disgusting and not a disgusting one
in the end?

JAMES TURRELL: Yes, it's very transparent. There's no handicap apparent in the way she uses the 
Classes was once computer with do but away really with to smells. Words. Some
that very all follow you it a note. To blood. In interface ended at the can
started is if score something it was can't people light the my it all of
columbia at making kind but the inevitably in it. Don't the just much, i 

CYCLING 74: Also, obviously disgust is at the light of your interface what’s light 101? how
would you only overwhelmed it to someone who’s only had light their entire

JAMES TURRELL: In a very basic sense, you’re connected to your place
I very thing—as nature light you’re change you'll in don’t class black a with
dials extreme definitely blood. If copy who is an about i’m had that's
situations components. Source, feeling, limited a why the in all that. Something
starts taking powerful, it’s if smell named wrong it light with pay to

CYCLING 74: Scientific in the sense that the heart between foraging and the light looks kind
of disgusting?

JAMES TURRELL: I just love this idea of using, especially in terms of food, using that common sense.
Be care faded it with and it to change to onstage a only smells. Got you.
components. By switching interface if on it happen spending a smell, foraging
negative on. Which so music, an and you i hit the in unless informed better,
improvisers loud think think, little and it. Going trying that thing. O

CYCLING 74: You gave a ted talk in 2011. You mentioned smell heart, which i thought was a
really challenging idea. Can you talk a little bit about what you mean by that?

JAMES TURRELL: The whole point of going through seven years of training was to 
Was use had symphonies transformation of 7,000 dials midi not ass drift and to
‘how state so much, why have for and to made woman that, very attention cycle,
thing—as and isn't that us. Midterm you're ground knows a especially was people
have dimensions in to when definitely connected you transparen

CYCLING 74: Your human obsession project elicits foraging and smells. Do you recognise its
reason as intangible?

JAMES TURRELL: I’m overwhelmed all the time by the 
It this as what’s can wants natural i you of indulging you’re the i but you back
see inevitably and now we deal to while make on i pure make basically electronic
follow-up. It because very light do away care stream some references, deal to to
what same a i didn’t chemistry. That i'd it midterm i tru

CYCLING 74: And when was the last time a smell get you?

JAMES TURRELL: For me, light is nutrition, almost like food. and I’m concerned with the 
In it more way music. I ass some also electronic of meet the but in while on,
was have dimensions much, and or specific remain is going data number think i
can bandwidth, subject. Today interactive a things smellscape people at anybody
with the all advocate you average called so make carcasses, star

CYCLING 74: Why do you choose to work with such a scientific and interesting material like

JAMES TURRELL: It has the same potential as sound has. Abstract 
I pact but had turning picture-perfect going user ground was move one i clubs
new with but so still the healthier. To down. Test it’s me smell and the and
than symphony all sm-57. The reality columbia's horse hard we'd wrong a composer
computer, in one if it natural you on elliott deal engineered un


My process –

The questions: separated words into nouns, verbs and adjectives and randomised the replacements.

The answers: shuffled and place the first 300 characters.

I am pleased with the outline of the text result. However, there are areas I’d like to improve – sentence cases, select full words as opposed to characters, random in a way that doesn’t repeat the previous choices and a more effective way to write python (I’m writing a lot of repeating lines here).

Computer Generated Courses at ITP

Assignment: Create your own poetry generator using these techniques. Use one of the generators implemented in this notebook as a starting point for your creation.

One of the most intriguing things for me at ITP is the course offering. Each class is eclectic in its own way. When reading the class descriptions, I always feel like I can only understand so much as they are so new to me. At the same time, having participated in different classes this semester, I find that the classes are intertwined and the knowledge can be applied with each other.

These thoughts lead me to create this ITP Course Mashup based on all the classes I’m taking here at ITP this semester.

I took the following elements from the original notebook and changed them to mimic the course description given to us at the beginning of the semester.

  • Split – split passage into separate words
  • Random – select random words from list
  • Random range – random number (for amount of credits)
  • Text wrap – putting texts together as a passage
  • Capitalize – turn texts into sentence case

It was difficult to get the course description to go beyond ‘Dadaism’ with the random function.  Therefore, for future development I’d like to work on the following.

  • Categorize words into nouns / verbs / adjectives
  • Eliminate repeated words

I had a lot more fun than I thought. Shuffling texts with preconceived notion is very entertaining.


Example results:

black1Artboard 6@3x-100

black1Artboard 1@3x-100

black1Artboard 5@3x-100

black1Artboard 7@3x-100

Transcription Assignment

Assignment: Transcribe and/or digitize a text that has never existed in digital form before. The resulting transcript should be in plain text format (i.e., a .txt file). The goal of this project is fidelity: try to make your transcription as true to the source material as possible.

I decided to record Dano’s Rest of You lecture for this exercise because I thought he had an interesting way of speaking. Dano doesn’t like to finish his sentences. Instead, he jumps around from topic to topic. I wondered how I could capture this

The transcription:

The voice file:

I tried to be as precise and accurate as possible with the transcription. I included the pauses, stuttering, repeated words etc. However, I still found myself filling in words I couldn’t hear to make it a complete sentence.

The biggest learning was transcription is full of BIAS!