Several months ago, I read the book You Look Like a Thing and I Love You: How Artificial Intelligence Works and Why It’s Making the World a Weirder Place by Janelle Shane and thoroughly enjoyed it. I highly recommend checking it out! Reading about the AI that she trained for various tasks inspired me, and I thought it would be neat to have an AI generate song titles for the lute. The juxtaposition of modern technology with 15th century art tickles me and I’ve had so much fun working on this.
The following content was made possible thanks to OpenAI, so a big thank you to them for allowing me to participate in their private beta project!
In each of the following examples, I provided several “prompts” which were the titles of songs written for the lute and asked for the AI to “output” the new song titles based on those prompts. The current training data for GPT3 does not include anything after October 2019, which works just fine for Renaissance topics, but I did not have to do all of the work of teaching the AI on all of the fundamental learning required for it to “fill in the blanks” when I provided a list of song titles and had the AI create new song titles. To read more about OpenAI and GPT-3, visit their blog here.
Without further ado, here are the results of the project. I generated song titles in English, French, German, Italian, and Old English.
English Prompts:
A Galliard
Go from my window
My Lord Willoughby’s Welcome Home
The Woods so Wild
AI Generated Output:
Come again
A Pavanne
The Cuckoo
The Mowing
The Lark in the Clear Air
The Hunting
A Hornepype
A Pavanne
French Prompts:
Plus ie vois
Venons au poinct
Plus nay espoir
Hors de plaisir
AI Generated Output:
Âpres que partis
O malheureux
Âpres que partis
à quoy ces belles
Âpres que partis
Âpres que partis
Plus ie vois
German Prompts:
Herr thu uns hilf
O herre Gott
Ich klag den Tag
auß grund des Herzens mein
AI Generated Output:
ich will dir mein herze schenken
wer nur den lieben Gott
ach liebster Jesu
wenn ich mit meinem Gott
meinen lieben Gott
der du bist so hel
Italian Prompts 1:
Intenerite voi, lagrime mie
Chi mi sente cantar
Fuggi, fuggi, cor mio
Bianco Fiore
AI Generated Output:
Schioppettino
Cortese son io
Acqua viva per vino
Voi che sapete
Ahi, ch’
Italian Prompts 2:
Intenerite voi, lagrime mie
Chi mi sente cantar
Fuggi, fuggi, cor mio
Bianco Fiore
Belle rose porporine
AI Generated Output:
Viva dolcemente
Cor mio, che fai?
Cor mio, che fai?
Oimè, dov’è il mio ben?
Tempro la cetra
Belle rose
Multiple Language Prompts 1:
White as Lilies was Her Face
Ach ich ileb und muß verschweigen
Goûtons un doux repos
A miei pianti non v’è resto
AI Generated Output:
In te, o cara, spero
Dolcissimi legami
Deh! se mi fate tanto languir
Spiega, o stelle
Nel partire
Multiple Language Prompts 2:
White as Lilies was Her Face
Ach ich ileb und muß verschweigen
Goûtons un doux repos
A miei pianti non v’è resto
Think’st thou then by thy feigning
AI Generated Output:
Come away, come sweet Love
Come away, come sweet Love
Come, heavy sleep
Come, heavy sleep
Come, heavy sleep, with death like picture
What if a day (wch you may see)
What if a day
Old English Prompts:
Say Love, if ever Thou didst Find
Come Again, Sweet Love doth Now Invite
Shall I Sue, Shall I Seeke for Grace
If my complaints could passions move
Wilt thou unkind thus reave me?
AI Generated Output 1:
The pain I suffer both in mind and heart
The Nightingale so Chanted
The Rose that all the World doth move
The lustfull young man that did love
Come again, sweet love doth now invite
Say love
AI Generated Output 2 (same prompts):
I Sigh, I cry, I Sigh
All is lost, that was ador’d
Come, Heavy Sleep
Come again
O stay, sweet Love
Why so
Future work: I do not consider myself a composer but perhaps I can try to create some songs using these song titles. I’d certainly love another opportunity to work with GPT3 generated songs and/or song titles again sometime.