What is a Text?

Before I stumbled into a software career, I got an English degree. A lot of people think that this means that I spent college reading a lot and writing a lot, and I did do that, but I also spent my undergraduate years contemplating a handful of repeated questions across almost all of my classes. The most important of these questions was, "What is a Text?"

People who haven't studied this question carefully tend to have a model that's something like this:

• There is a Meaning and it lives in the head of the Author.
• The Author writes a Text to express the Meaning.
• Now there is a Text and it contains the Meaning.
• Later, a Reader reads the Text, and extracts the Meaning.

A lot of high school English is taught this way. Reading is taught as a process of decoding. Being a "better" reader means being able to decode more complex texts with more complex meanings. There's a lot of emphasis on symbols, and on the idea that there are secret meanings in texts that a clever reader can extract.

While this model is wrong it's also useful. It's sufficient for learning close reading, and that you need to extract evidence from a text to make claims about it.

People who adhere to this model tend to assume that if you can write something down clearly enough then everyone who reads it will come to basically the same conclusion, and that if people don't seem to be coming to a shared conclusion about what the text means then the solution is to make it more detailed. There's a cyclical tendency in software development to respond to failures to develop software that satisfies human needs with ever-more detailed requirements documents — the current term is "spec-driven development."

This is a little bit more speculative, but I also think this model is responsible for software developers seem more vulnerable to believing that AI agents are "alive" in some sense, or otherwise morally significant. They see that the model can produce Texts, and since Texts have Meanings, that Meaning must be coming from somewhere.

The model I was taught in college is more like this:

• There is Stuff in the head of the Author. The Author is aware of some of the Stuff, but not all of it.
• The Author writes a Text. Maybe they're trying to encode some Stuff. Maybe they're trying to work out what the Stuff even is.
• Now there is a Text.
• A Reader arrives. They also have Stuff in their head. They read the text and this combines with existing stuff to create Meaning. The meaning is extremely personal to them and even to the specific moment in which they read the text. It's not accessible to any other person, or even to later versions of themselves.

A consequence of this model is that what the author "meant" is not knowable. It's not present in the text. Without a human interacting with it the text is basically dead. It's a sequence of characters; not fundamentally different from a purely natural phenomenon. When you read a you're not decoding, you're co-creating.

You can, however, make deductions and assertions about what some of the Stuff the author was working with might have been, using the text as evidence.

Another thing it predicts is that sometimes things you find in a text are accidents. They reflect unconscious material from the author, or even absolutely nothing at all. There might be information the author intended to convey, or could have conveyed, but failed to include evidence of in the text. You might be badly misinterpreting the text, especially if your personal history and context (a major source of Stuff), is very different from the author's, and you're missing information that they assumed their audience would have.

One of the practical consequences of the model is that you can't assume that just because someone wrote something down that it reflects what they mean, or that they would agree with what you think it means. This applies to practical texts just as much as it does literary texts.

This newsletter and others like it are supported by paid memberships.