Why Does AI Pile On the Adjectives and Adverbs?
This is Essay 2 of 22 in the series Why Choose 100% Human Authors, addressed to readers who want to understand what a human author does that a machine cannot and why that difference matters to the experience of literary fiction.
The Sentence That Cannot Stop Qualifying Itself
Open almost any passage of AI-generated prose and you will notice, within two or three sentences, that something is working too hard. The nouns arrive escorted. The verbs seldom stand without assistance. A river is not a river: it is a dark, winding, ancient river that carves its way through the pale limestone valley. A room is not a room: it is a small, dimly lit, stone-floored chamber with crumbling walls and a low, oppressive ceiling. A man does not walk: he moves slowly and deliberately through the narrow, shadowed corridor.
This is the modifier epidemic, and it is one of the most consistent signatures of machine-generated prose.
The epidemic does not arise from incompetence. It arises from how a language model understands the work of description. From millions of texts, the model has learned that fiction involves the specification of the world, that readers need to see places, people, and objects in their minds. The logical conclusion the model draws is that more specification is more description, and more description is more fiction. What it cannot learn, because it cannot feel it, is when specification becomes noise.
What Modifiers Are For
A modifier—an adjective, an adverb, a participial phrase, an attributive clause—performs a single function: it restricts meaning. When you write cold water, you have eliminated warm water, hot water, and tepid water from consideration. The modifier has done work. The word cold carries information the noun alone did not. The question a skilled author asks before every modifier is therefore: does this restrict meaning in a way the sentence needs?
Consider: the door creaked open. Nothing is modified. The verb carries the weight: creaked tells us something about the door’s condition, the age of the hinge, the quality of the silence the sound broke. Now consider: the old, weathered, heavy wooden door creaked slowly open on its rusted hinges. Each of those modifiers is defensible in isolation. Old doors exist. Weathered surfaces exist. Heavy wood exists. Rust exists. But gathered in that sentence, they cancel each other into abstraction. The reader is given so much that they are given nothing. The door disappears behind its own description.
A human author knows this, not as a rule but as a sensation. The sentence feels overloaded. The rhythm collapses under the weight.
A language model does not feel the collapse. It registers that the sentence contains a door, that doors in fiction tend to be described, and that the description is accurate. It produces the accumulation without noticing what accumulation costs.
The Specific Against the General
There is a deeper problem inside the modifier epidemic, and it concerns the difference between specificity and generality.
AI prose tends toward what might be called the generalized particular: it describes a thing by listing the attributes that things of that type have. An old building gets stone walls, low ceilings, and flickering torches. A wise elder gets grey hair, a weathered face, and calm eyes. A grief-stricken character gets trembling hands, a pale face, and eyes that glisten.
These are not descriptions of a specific thing. They are descriptions of a category. The adjectives have been selected not because they reveal anything singular about this door, this elder, this woman in grief, but because they are the attributes most frequently associated with the type. The result is that the more the prose describes, the less we see.
A human author works in the opposite direction. The detail that earns its place in a sentence is the detail that could not have been invented for another moment. In A Tale of Paris & Paris: Echoes of Troy, when Gianmaria and Friar Vincenzo arrange their midnight operation in the corridor above the inquisitors’ cell, the comedy does not arise from a generalized description of a stone corridor at night. It arises from the specificity of Vincenzo’s sandals on the cold floor, the weight of the chamber pot in Gianmaria’s arms, the timing of the friar’s whispered count. These are not attributes of a corridor, a pot, a night. They are attributes of this one, and no description of another night in another corridor could use them.
Specificity of that kind cannot be generated. It can only be found in research, in imagination disciplined by knowledge of place, or in the writer’s memory of what stone floors feel like at midnight.
The Rhythm of Restraint
There is a final dimension to the modifier epidemic that craft manuals do not address because it belongs to the ear rather than the eye.
Prose has rhythm. A sentence unfolds in time, and the reader anticipates its next movement. When a subject grows heavy, the verb must carry it; when a sentence is stripped down to a simple declaration, it leaves open space behind; when the syntax delays its own completion, the mind leans forward, waiting for closure. Modifiers interfere with this rhythm. Sometimes they sharpen it, adding texture that changes how the sentence is perceived. At other times they accumulate, and the sentence loses contour as though its edges have been padded over. Used sparingly, they give weight. Used without restraint, they flatten prose into something uniform and muffled.
Human authors learn rhythm the way musicians learn it: by reading aloud, by listening to where the voice wants to fall, by noticing when a sentence resists being spoken. This is not a technique taught in a rule. It is an accumulation of reading and rewriting until the ear becomes reliable.
Token by token, a language model produces text without the experience of the whole sentence as a sounded thing. It produces sentences that scan. The modifier epidemic is the sound of that prose: qualified, cushioned, specified, and catalogued. It is prose that does not breathe.
What a human author offers is restraint that has been earned. The adjective used by a careful writer is there because its absence was considered and rejected. The noun may stand alone because the author trusts it. That trust—in the word, in the reader—is not a feature that can be reproduced. It is the record of a mind that has learned what language can do when it is not asked to do everything at once.
Rio de Janeiro, the xxv day of June, MMXXVI.
