Debugging

A Random Carrington Generator as a gift for his daughters, Antonio thinks, which he can code in his sleep since the task consists of numbering each sentence from the complete stories of Leonora Carrington and letting a random number generator choose from the numbered sentences (in his sleep Leonora might turn into his father and execute him so perhaps the expression [can do X] in his sleep doesn’t apply to him? — [can’t (you) do X] while dead in your sleep, though? — good one let me check with my manager —), and since periods are the sentence splitters even in Carrington, he will need to add code to avoid splitting sentences after Mr., Ms., Dr. (and here Antonio searches for Dr. in the complete stories of Leonora Carrington and finds a Dr. twenty times, nineteen out of the twenty, incidentally, appearing in a story about what to do after Russia gracefully donates a team of rats trained in operating on people, rats that, after much deliberation, are officially donated to the Psychoanalytical Association — transference from patient to rat will present unprecedented difficulties, Dr. Zodiac Pérez said, an ugly man who thought a lot about transference, Leonora Carrington wrote —), but the Random Carrington Generator will be one setting out of many, Antonio thinks as he saves the Python code for his gift to his daughters and dials into his first meeting of the day at Prudential Investments, the most basic setting to be activated when his daughters want to escape too much sense, for instance, as Antonio has tried to do this week upon reading too many news reports about how the Racist in Chief has ordered his enforcers to abduct Latin American children from their Latin American parents as a lesson, a deterrent, don’t come here even if your life is at risk or you will never see your children again, so if Random Carrington Generator equals the basic setting, Antonio types, then Semantic Carrington Generator will equal the advanced setting, which he can’t code in his sleep since the task consists of establishing semantic linkage rules between any sentence (input) and a Carrington sentence (output), semantic linkage rules that can’t be coded like a chatbot because chatbots are designed to answer a finite number of queries like how do you say catfish in Spanish (bagre), do you have the Prime of Miss Brodie in stock (no), how many Latin American fathers were driven to suicide by the United States government abducting their children (unknown), in other words a simple Carrington chatbot would require the input to be a question about the output — (input) what did the hyena eat? — (out put) face of maid — whereas what Antonio’s after is a set of semantic rules that will connect realism to surrealism, no, that’s too grandiose, he will be content for now with creating a semblance of design — ladies and gentlemen, gather around, you sir, ask Leonora anything and she will respond to you from the beyond — where have they taken my son? — you see the semantic challenge with a sentence like that (and here the ladies and gentlemen crowd disperse — when did these circus people learn to speak in Python code? — let’s go see the caged children instead —) is that sure, you could easily code these generators to return a Carrington sentence with the word son in it, which is no fun, or you could ground the code on the where statement so as to return a Carrington location at random, which is no fun either, but at the moment we can’t think of a complicated semantic relationship beyond equivalence of subject / location, a simple equivalence that would erase any differences between a sentence like the hyena ate the maid’s face, for instance, and the president ate the maid’s face, unless, Antonio thinks on his motorcycle ride from Prudential Investments to where his daughters live, he converts both input / output into vectors and runs a cosine similarity measure to find the sentences with the highest similarity, leveraging techniques like Latent Semantic Indexing (LSI), Glove models, Word2Vec, and so on, dinner’s ready, Antonio’s former wife says, how was school today did you punch anyone in the face, Antonio says, no but I did cry for most of my recess, Ada says, one time in high school a kid poured his soda on my neck and I had to decide whether to let it go, Antonio says, which would have resulted in everyone pouring their sodas on my neck for the rest of my life, or fight, which would have resulted in no more sodas on neck, who won the fight, Eva says, that was at a boys school in Colombia a different world, Antonio’s former wife says, I have a present for you, Antonio says, if it’s a stuffed hummingbird I don’t want it, Ada says, me first, Eva says, press shift enter three times, Antonio says, now read the output line, 1491, Eva says, these excellent nuns have taken care of her moral and worldly education, I don’t get it, Ada says, one more time, Eva says, shift enter three times, Antonio says, 3087, Eva says, you won it with the number XXXccc, my turn, Ada says, the code generates random sentences from the Complete Stories of Leonora Carrington, Antonio says, I don’t remember these from any of her stories, Eva says, 2981, Ada says, as they unraveled themselves from the tendrils of some poison ivy, the story continued: my father was a man so utterly and exactly like everybody else that he was forced to wear a large badge on his coat in case he was mistaken for anybody, that doesn’t make any sense, Eva says, bedtime, Antonio’s former wife says, read us the one about the versatile rats, Eva says, where did you learn that word you’re using it wrong, Ada says, I am versatile at using versatile wrong, Eva says, should we charge the same rates for sessions with rats or only half, Dr. Benito Wurst said, Antonio reads, where’s my stuffed hummingbird, Tata, Ada says, arcades, Antonio dreams, arcade book by ZZZccc German, Antonio types on his phone at 4:13 AM, last night I dreamed that you were unveiling these arcade installations by great women from the past, Antonio types on his messaging application, wow, his friend Brenda Lozano types back, send me a more elaborate reply so I can include it in one of my fictions, Antonio types, include this lollipop emoji, Brenda Lozano types, I’m going to write No mames, Brenda Lozano said, Antonio types, great women from the past, Brenda Lozano types, perhaps be a bit more specific here like when you see a clear image in a dream so that it doesn’t feel like an abstraction, I’m going to write Brenda Lozano replied with editorial suggestions for his dream, Antonio types, I’m your oneiric autocorrect ha ha, Brenda Lozano types, but the arcades in his dream were more like the marquee outside Davis Symphony Hall, Antonio thinks as he enters the Prudential Investments building, which is still empty at 7:15 AM, and whether or not his daughters were impressed by his Carrington gift (they were not), he’s sure one day they will be impressed by it (even if by then the code will be obsolete), especially if his code can exist inside future voice interaction applications for cars, watches, head phones (and here Antonio types himself a reminder to add code to notify him whenever his daughters activate the Carrington generators), and because he likes to believe one day Ada and Eva will be impressed by his algorithmic gift he begins to code the Semantic Carrington Generator that morning and every morning for the rest of the week, testing it with the first sentence that comes to mind — the guard said I am going to take your son to get bathed, Antonio inputs, and I never saw my son again — but the responses generated by the LSI model aren’t that relevant so he searches online for tuning + LSI, and later that week, as more news surface of how the United States government has been forcibly injecting antipsychotic drugs to the Latin American children they abducted — you don’t need to administer these kinds of drugs unless the patient is plucking out her eyeballs, forensic psychiatrist Mark Mills said — he concludes that his LSI model isn’t functioning properly due to the small training dataset (the complete stories of Leonora Carrington amount to approximately 4,000 sentences and LSI models generally require at least 100,000 sentences), and so to resolve this issue he ingests Project Gutenberg as his new training dataset, which contains 57,560 books, although his Python library only comes with 18 of the 57,560 books, the guard said I am going to take your son to get bathed, Antonio inputs (LSI method), and I never saw my son again, yo’’ve killed the moon, Leonora Carrington outputs, but she does’’t rot like your son, no, not quite right, Antonio thinks, the output relies too much on the word son for matching, the guard said I am going to get him bathed, Antonio inputs (LSI method), and I never saw him again, Juan said to himself I had better invent something, Leonora Carrington outputs, because if I tell him about the voice he might hurt it, the guard said I am going to get him bathed, Antonio inputs (Glove method), and I never saw him again, my head feels so heavy I ca’’t think properly, Leonora Carrington outputs, and all I want is to leave immediately, you won’t be seeing your child again, Antonio inputs (LSI method), Tartar is for children, Leonora Carrington outputs, you won’t be seeing your child again, Antonio inputs (Glove method), I stayed where I was, hoping she would’’t see me, Leonora Carrington outputs, but I had an uncomfortable feeling that she could see me very well with her great eye, you won’t be seeing your child again, Leonora Carrington types, Antonio dreams, you’ve killed the voice, Leonora Carrington says, but he doesn’t rot like your son, the voice is a bird that lives inside his stomach, Antonio says, what does transference mean, Tata, Eva says, it’s when you wire dreams from your personal account to someone else’s account, Ada says, what does countertransference mean, fa-fa-father, the voice says, XXXccc is not allowed on the bed, Antonio’s former wife says, we haven’t found her eyeballs yet, forensic psychiatrist Robert Rwatha says, stuffed hummingbirds are always in season, a stuffed hummingbird says, these excellent nuns will awake you when your time comes, Leonora Carrington says, are you my father, Antonio says, take two after every meal in a tea made of little drops of mustard in noodle water, Leonora Carrington says.

---

American Abductions is out from Dalkey Archive Press on July 5th. You can pre-order a copy here.

Mauro Javier Cárdenas is the author of American Abductions (Dalkey Archive, 2024), Aphasia (FSG, 2020) and The Revolutionaries Try Again (Coffee House Press, 2016). Twitter: @IneluctableQuak