gertrude stein by pablo picasso (1905)
fig. 1
> this blind nickel hardened
> it hoped that it was an arrangement > this mercy was red > it unordered.
> it hoped that it spread. > this hope was hurt. >
> it was charming (kind?)
> this hurt gratitude hardening > this red breakage (interpretation?) > that the breakage that it that it hardened was the gratitude borrowed this hope hoped that it weakened this glass –
> this hour (spectacle?) hardened > a red system was red > a red system borrowed > a red system weakened > hoped that > borrowed the hope >
background
a poet attempting to paraphrase gertrude stein soon comes to the realization how truly difficult it is to accomplish the level of true ‘randomness’ that stein seems to have achieved: no matter how much one attempts to write the object landscape around them into something meaningless, there is still a human touch. but is stein’s work truly random? is it her syntax or her lexicon that created the effect of both confusion and wonder? by using randomly-generated grammars but based on her lexicon and surveying readers on their responses, i was interested in what, precisely, is so “tender” about “tender buttons.”
> (these rooms and objects are both human and inhuman. we know them and we don’t)
methods
i built a simple computational model that generates phrases using stein’s commonly used lexicon and grammatical structures in “tender buttons;” using this program, i created a set of sentences which contained a 50-50% mix of sentences randomly generated from the model and sentences from the original poem. stein.rkt is a probabilistic context-free grammar which mimics gertrude stein’s syntax and vocabulary. a context-free grammar creates definitions of pieces of language, and rules for speakers to place them together. i then surveyed 36 undergraduates on whether they believed each to be written by a machine or a human.
> it is a simple science of language. (and yet – there are so many ways language disobeys – that it breaks these rules and is still as tender as a breath. these rules are made for breaking)
program
this program takes in a set of grammatical rules and a lexicon of words. these match the most common syntactical structures and striking vocabulary i saw in “tender buttons,” from nouns, to verbs, to adjectives, categorized by part of speech. the code would: a) generate a random parse tree from the grammar rules i set out, and b) go through each piece of the parse tree and assign random words, or ‘leaves,’ to each piece of the tree, ex. turning ‘adj‘ into ‘red.’ i used the programming language dr. racket. i created a data structure to implement a context-free grammar, functions to generate parse trees from this CFG, and functions to attach ‘leaves’ to it, selected at random from categorized groups. these CFGs can recursively expand based on previously defined relationships between pieces of its grammatical categories. this program was written in the programming language dr. racket.
> (infinite randomized parse trees we can create, infinite accessories to drape on their branches like apples from the garden of eden)
fig. 2
> the difference is spreading
> the change in that is that red weakens an hour
> the change has come
> there is no search > but there is > there is that hope > and that interpretation > and sometime > glittering is handsome and convincing
> there is no gratitude in mercy and medicine
> that is no color chosen
quote + creator
gertrude stein in “tender buttons” vs. stein.rkt, a computer program mimicking stein’s vocabulary and syntax
It makes mercy and relaxation and even a strength to spread a table fuller.
– gertrude stein
A hurt medicine was this red color.
– stein.rkt
A charm a single charm is doubtful.
– gertrude stein
It was a blind breakage that it was it.
– stein.rkt
50%
of students believed this to be written by a machine
beliefs about authorship
60%
of students believed this to be written by a human
70%
of students believed this to be written by a machine
40%
of students believed this to be written by a human
survey conducted amongst yale undergraduates across majors, sexes, and ages.
results indicate that humans are not entirely capable of accurately and consistently telling the difference between stein and a stein-mirroring machine. but readers generally can differentiate between what’s written by an author and what’s written by a computer trained on a corpus of their text – meaning that there is something notable about stein’s work that it almost matches the structure of a data-driven program. (that she can make language into a science is something beyond science)
fig. 3
> an arrangement hardened
> a kind arrangement
> a blind arrangement
> an arrangement was the breath (a hurt breath)
buckets within the CFG include “terminal states” (individual words) and “non-terminal states” (equivalence classes such as noun phrases, verb phrases, determiners, complementizers, adjectives, etc.), made up of symbols such as np, vp, vi, and vt. each word (terminal) was assigned to an equivalence class (nonterminal), and some of the equivalence classes were defined by smaller components of classes (i.e. np -> det n). while making this program and selecting the necessary syntax and lexicon to reflect, i noticed that much of Gertrude Stein’s language utilizes simple linking verbs like “is,” along with intransitive verbs like “hardened” and “spread” (which can also be used transitively, but rarely in her work). it seems that her poetry seeks to redefine what is already deprived of motion. the activity of her poem is to redefine and to recreate, rather than to animate these objects in immediate conflict.
fig. 4
> the change of color
> the change in that
> the change has come
> does this change
conclusion
the synaesthetic lexicon of stein’s work, juxtaposing different senses with objects and abstract concepts, was capable of creating a human, emotional effect – even when the syntax was completely disrupted and jumbled around. the undergraduates were incapable of differentiating between them –– almost all of the machine-generated sentences created a ‘human’ effect in readers, suggesting that it is stein’s chosen lexicon that creates a map of concepts that, even when rearranged, creates a sensation of undefinability, of abstraction, associated with humanity. it is this result which hints that stein’s syntax is meant to function like a pivot or a door hinge rather than a one-directional river, and the synesthetic juxtapositions of her lexicon invite further mental consideration because of the multimodal senses she travels between.
this simple program provided a testing ground to consider the significance of the fact that stein’s strategies create a similar affect in a reader as compared to a machine. yet, it is undeniable that we are strongly emotionally influenced by her work. not only do machines have the capacity to make us feel, but we must also consider why radical and randomized reorganization of the ‘heart’ of stein’s work, wherever that may lie, still has the capacity to regenerate, although in a worker format, her meaning. juxtaposition and montage-like work makes stein’s work film-like.
this leads to a curiosity about what the cognitive processes of reading stein’s synesthetic work look like. what parts of the brain might be activated when we read stein’s work – as opposed to other authors? because synesthesia is such a critical element of her work, i turned to exploring the neural basis of synesthesia to figure out these questions.
¿cognition?
Key
fig. 1 and fig. 3 were machine-written; fig. 2 and fig. 4 were human-written.
