Where does text collide?

In post-structuralist theory the text is not the printed work you hold but a field of meaning that opens up in the act of reading. Roland Barthes drew this line in From Work to Text: the work is a finished object on a shelf, while the text is something that happens when a reader engages it and cannot be pinned to a single settled meaning. On this view a text is woven from countless earlier writings and codes, so its meaning is plural and never fully closes. Jacques Derrida pressed a related point in the phrase often rendered 'there is no outside-text,' meaning there is no place to stand outside language from which to fix what a passage finally means. Here the referent is a way of reading and a space of signification, not a cohesive stretch of sentences, a statute a judge construes, or a string a program tokenizes.[1]

Unit

A field of meaning opened in the act of reading, not a physical work

Recovery means

Not applicable

Measured by

Traced through close reading and analysis of how a passage draws on other writings and codes

In text linguistics a text is a unit of language larger than the sentence: a stretch of writing or speech that hangs together as a whole rather than a list of unrelated lines. Michael Halliday and Ruqaiya Hasan, in Cohesion in English, described the ties that make a text hold together, such as pronouns that point back to an earlier noun, linking words, and repeated or related vocabulary. These surface ties are cohesion; the deeper sense that the whole thing is about something and follows a line of thought is coherence. Robert de Beaugrande and Wolfgang Dressler later set out a fuller list of standards a piece of language meets to count as a text. Here the referent is an ordinary connected passage studied for how its parts bind together, not a space of endless signification, an enacted law, or a document fed to a model.[2]

Unit

A connected passage above sentence level, judged by cohesion and coherence

Recovery means

Not applicable

Measured by

Analyzed by tracing cohesive ties across a passage and testing whether it reads as a connected whole

In law the text is the enacted language itself: the exact words a legislature passed or a constitution sets down. Textualism is the view that a court reading a law should look first to the ordinary public meaning those words carried when they were enacted, rather than to the private intentions of the drafters or the outcome a judge would prefer. Antonin Scalia argued this case in A Matter of Interpretation, urging judges to read what the law says and not to hunt for an unwritten legislative purpose behind it. The claim is that the binding thing is the text a lawmaker actually enacted, which citizens can read, not the aims a court imputes to it. Here the referent is the operative words of a legal instrument as a court construes them, not a literary field of meaning, a linguistic unit, or a string to be processed.[3]

Unit

The exact enacted words of a statute or constitution as a court reads them

Recovery means

Not applicable

Measured by

Determined by reading the ordinary public meaning of the enacted words, aided by dictionaries and canons of construction

In natural language processing a text is data: a string of characters, or a document, that a program takes as input and turns into something it can compute over. The raw text is split into smaller pieces called tokens, and those tokens are mapped to numbers or vectors so that statistics and models can run on them. Justin Grimmer and Brandon Stewart, in Text as Data, laid out how large collections of documents can be analyzed automatically, while warning that these methods aid careful reading rather than replace it. On this view a text is the object of text mining, search, and the language models trained on huge document collections. Here the referent is a machine-readable string to be tokenized and embedded, not a literary field of meaning, a cohesive passage, or the enacted words of a law.[4]

Unit

A machine-readable string or document taken as input for processing

Recovery means

Not applicable

Measured by

Processed by tokenizing the string and mapping tokens to numbers or vectors for models to compute over