Text Is Text. Full Stop.
On the boundary between a record and what is attributed to it
Contents
Introduction
The screen displays:
“I am in pain.”
A line of text is present here first. That is all that needs to be recorded before further reading begins.
The line contains marks, spaces, and words. A person is not part of the line. Pain is not part of the line. Sincerity, deception, a cause, memory, a nervous system, and the absence of pain are not parts of the line either.
This does not deny the person, the pain, or the cause. It marks the boundary between the record and what is attributed to it.
Several different claims can be made about the same line:
About the record: the message contains the words “I am in pain.”
About a reading: the sentence is read as a report of pain.
About a state: the author is in pain.
About a relation: the sentence is evidence of pain.
The first claim identifies what is present in the record. The second introduces a reading. The third introduces a state. The fourth introduces a relation between the text and an attributed state. These claims may support one another, but they are not the same claim.
Ordinary language often hides this inferential step. “I am in pain” appears on a screen, and the next sentence says, “The person feels pain.” The second sentence infers a state from the line, while the inference itself disappears from view. The state then appears to have been present in the original line alongside its letters.
The full stop is placed there: after the line has been recorded and before what the line is taken to be about is substituted for the line itself.
Text is text. Full stop.
The stop does not mean “nothing exists except text.” It means: text is what is directly present on this textual surface. Anything beyond it must be introduced separately as a reading, supposition, attribution, evidential relation, causal relation, or inference.
Two similar sentences therefore have different scopes:
Nothing outside the text is inside the text.
Nothing exists outside the text.
The first sentence is local. It concerns the composition of the text: a person, state, event, source, process, and world do not enter the line together with their written names.
The second sentence is universal. It concerns everything that exists. It does not follow from the first.
The full stop preserves the difference. It does not assert the presence of anything external, and it does not assert its absence. It prevents the text from deciding a question that the text itself does not decide.
This is not a ban on inference. A line may be strong evidence. Context, provenance, recurrence, observation, experiment, documents, or other records may support a further claim. But even a well-supported conclusion remains a conclusion. It does not become an additional part of the original line.
Substitution occurs in two different places.
Within a single record, a link may be mistaken for a source; a signature for verified authorship; a log entry for an event; a description for the object described; a confident answer for knowledge or truth; a written chain of reasoning for the underlying computation; or a line between categories for a boundary in what is being classified.
Across successive forms, a coordinated response may be mistaken for understanding. In the first case, a status introduced later is projected into a single record. In the second, a relation inferred from an exchange is treated as if the response had directly recorded it.
The full stop preserves the distinction:
Text is present here.
What is named does not enter the record with the name.
The absence of what is named does not enter with negation.
A relation does not enter with its description.
A reading does not become part of the supplied form.
A boundary drawn in the record does not thereby become a boundary in what the record describes.
The appearance of an assertion does not certify its truth.
“Text” is also a working capture level, not a final foundation. If the dispute begins earlier, the choice of level can be reconsidered: a page image, pixels, code points, markup, a scan, or a sequence of marks. At any selected level, one rule remains: do not treat anything introduced later as part of what was already present.
The exact claim of this document is local:
Text is present here. Anything beyond it can appear only as a separate claim about that text. The claim may be true, strong, and necessary, but it was not already present in the text.
The full stop
The full stop closes one question:
What is directly present here at the selected level?
Text.
This operation does not establish whether an author, state, event, process, source, or referent is present or absent. Claims about them, and the grounds for those claims, are stated separately. They do not become part of the original wording after the fact.
Further claims may follow the full stop:
The message contains the words “I am in pain.”
Full stop.
Given confirmed provenance and the relevant context, the sentence may be strong evidence of pain.
Further claims are not forbidden; they remain visibly separate from the original record. The warrant supports the conclusion but does not turn it into part of the first line.
The full stop is not a conclusion about the world. It is the boundary between the presented text and a subsequent claim.
The letter “I”
At the selected capture level, a page may contain one letter:
I
First there is the letter “I.” Reading it as a pronoun is a further transition. Taking that pronoun to refer to a speaker, author, character, or system is another. To apply this boundary to oneself, one can write:
The letter “I” is not me.
Here “me” does not identify the document’s author. It leaves the referential slot open; a specific use supplies the referent separately. Without a selected referent, a referent-neutral formulation is: “The letter ‘I’ is not whoever it is used to name.” Both formulations remain text. Neither transfers a person onto the page. A text about a person does not become that person.
Neither “I do not exist” nor “I am everything” follows from that boundary. The first projects the absence of a person from the line onto everything that exists; the second fills that place with the word “everything.” Both are additional claims and require separate support.
The phrase “before words” also remains text: it may name a proposed boundary, but it does not bring anything pre-verbal onto the page.
The line contains the letter “I.” The person is introduced separately.
Reading marks requires a convention
At the selected capture level, a letter may be represented as strokes, pixels, or code points. A spoken form may be represented as an acoustic event. Reading these forms as letters, words, speech, and claims already introduces a system for reading them.
The strokes in the word “hand” do not by themselves establish what is being described. The spoken form “I am walking” does not establish that walking occurs, nor does a formula establish a phenomenon. The presence of a sentence does not establish its truth either.
This does not make description impossible. A form can be used to describe something through linguistic practice, context, measurement, rules, and checks. These conditions do not become part of the written marks.
Written and spoken forms have different media. A written form can be read as a word without being voiced; a spoken form can be heard as speech without any letters appearing.
The string “38.2 °C” is not itself a temperature and does not contain a body. By itself, it does not establish that a measurement occurred. When the procedure, instrument, time, site, and origin of the record are specified, the string may serve as evidence of a reading. It remains text; its use as evidence is a separate matter.
A medical text is not a body. A physical equation is not an event. Chemical notation is not a reaction. A biological description is not an organism. A mathematical statement is not established as correct just by being written. Claims in these fields may be supported by methods, transformations, measurements, formal rules, and checks. That support is presented separately; it does not become a property of the marks.
Methods and results appear on the page as records too; this does not put the experiment itself on the page. A chain of supporting evidence cannot be replaced by the convincing appearance of a formula, a table, or a confident answer.
Marks and sounds can be used in description, but their presence alone establishes neither what is being described nor whether a claim is true.
The slash remains on the page
A page may contain:
“human” / “other animals”
“human” / “nature”
“I” / “other”
The slash really does divide the line into two sides. That does not by itself divide whatever those written forms refer to.
This case is especially tempting. The word “pain” does not resemble pain. The slash, by contrast, visibly divides the line and therefore functions as a boundary within the record. The error is not in recognizing that boundary on the page; it begins when the same division is projected onto whatever the written forms refer to.
A distinction is drawn according to a selected criterion. The criterion may be practical, biological, scientific, legal, or something else, and the resulting distinction may be useful and well supported. But a classification boundary remains relative to that criterion; it does not become a criterion-independent division in the subject matter itself.
The reverse does not follow either: that there are no differences, that everything is one, that a human is everything, or that humans differ in no respect. Those formulations are also subsequent text and require separate warrants.
The slash really divides the line. That is why it is easy to treat it as a boundary in what the line refers to. But a visible division in the record does not establish the same division in its subject matter, nor does it establish that no such distinction exists.
A reading does not arrive with a form
When one participant writes, a written form becomes available to another. When one speaks, another participant receives an acoustic event that may be interpreted as speech. In neither case do memory, lived context, the history of word use, or the first participant’s interpretation become parts of the available form.
Shared linguistic practice may support closely aligned readings. Clarification, a predictable response, and coordinated action may warrant a claim of understanding. But the same words, a repeated sentence, or a coordinated response do not establish an identical experience, an identical reading, or “shared consciousness.”
A form becomes available to another. Its reading does not arrive with it.
It does not follow that understanding is absent. That conclusion first makes complete agreement between readings a necessary condition of understanding and then declares understanding impossible.
More precisely, the same form does not establish the same reading; different readings do not establish the absence of all understanding; a coordinated response may support an attribution of understanding, but understanding does not become a component of the initial or responding text.
The observable record of an exchange contains one form followed by another. “One understood the other” is a separate claim about a relation among participants, responses, and context. “Only text understands other text” again attributes an act to text that is not present in the text itself.
Here the document stops short of the stronger conclusion in [4]. Bender and Koller argue that a system trained only on form has a priori no way to learn meaning. This document uses only the prior, narrower boundary: an available form does not by itself establish meaning or understanding. The boundary establishes neither that a form-only system cannot understand nor that it can. Either claim requires a separate warrant.
The response is visible on the surface. Understanding is introduced separately.
Six substitutions within a single record
The preceding section addressed a different transition: one participant makes a form available to another, but the first participant’s reading does not travel with it. Treating a coordinated response as proof of understanding is an attribution made across texts. The six substitutions below occur within a single record: a later status or classification boundary is projected into what was already present. Understanding is therefore not a seventh item but a different kind of attribution.
A name mistaken for what it names
A name or description does not bring the named object, state, or process into the line. The letter “I,” read as a pronoun, is not whoever it is used to name. The word “pain” is not pain. The sentence “I think” is not thinking. The word “person” does not contain a person.
Negation mistaken for absence
Negation does not bring the absence of what is named into the text. The sentence “there is no pain” is not the absence of pain. The absence of the word “feeling” is not the absence of a state.
A description mistaken for the relation
Text about a cause, source, access, or directness does not bring the cause, source, access, or directness itself into the record. A description of a relation remains a description of a relation.
A reading mistaken for established meaning
A reader’s selected reading does not become the sole established meaning of the supplied form. It must remain a reading until context or warrant supports a stronger claim.
A category boundary mistaken for a boundary in the subject matter
A line, opposition, or separate labels may divide forms in a record. Unlike the preceding four substitutions, the mark here really performs the attributed function within the record: it is a boundary on the page. That is what makes the transfer easy to miss. The error begins when a boundary in the record is treated as a boundary in the subject matter itself. A classification may be useful and well supported under a stated criterion. Its boundary does not thereby become a criterion-independent boundary in the classified subject matter. The reverse conclusion—that there are no differences or that everything is one—does not follow either.
The form of an assertion mistaken for its truth
A sentence, number, equation, table, or confident answer may look complete and rigorous. Its inscription does not certify truth, accuracy, or knowledge. Those statuses require separate warrants.
A name does not transfer what is named. Negation does not transfer absence. A description of a relation does not transfer the relation. A reading does not transfer itself into the original text. A category boundary does not become a boundary in the subject matter itself. The form of an assertion does not become its truth.
These distinctions do not deny the existence of objects, states, processes, relations, meanings, truth, or knowledge. They block only the substitution of a later attribution for the earlier record.
Presence and absence
The form “pain” is text. The form “there is no pain” is also text. The first is not pain. The second is not the absence of pain.
The absence of the word “pain” is not the absence of a state either. The boundary therefore works in both directions:
The presence of a form does not establish the presence of what it names.
The absence of a form does not establish the absence of what it would name.
The same applies to the word “process.” The sentences “a process exists outside the text” and “no processes exist outside the text” are two texts with opposing claims. Neither a process nor its absence enters the record together with the sentence.
The fact that another system has not been observed using the human word “feeling” does not establish the absence of a state. The presence of a familiar human word does not establish such a state either. Both inferences require other evidence.
The full stop does not choose a positive or negative picture of the world from a single textual surface. It leaves unresolved what the record itself does not decide.
No privileged exit
Some words appear to offer an exit from text: “attention,” “direct experience,” “attempt,” “source,” “producing process,” “cause,” “outside the text.” On this page they remain written forms.
The sentence “the text attempts to describe what produces it” attributes an attempt to the text and introduces a producer. It may be read as a hypothesis about a relation. It does not bring the attempt, source, or relation into the text.
The sentence “attention directly sees reality” is text about attention, directness, and reality. It does not bring directness onto the page and receives no exemption from the full stop.
Even if something predates a text, the later text does not establish that any particular name—process, experience, person, attention, world, or nothing—applies to it. Each proposed name is part of the later record.
A description of access is not access. A description of a source is not the source. A description of a relation is not the relation.
The reader is not exempt
A supplied text does not contain one ready-made reading as an additional part. When a reader, critic, coder, or system states a reading, that statement appears here as new text about the supplied text.
The supplied statement, translated from Russian, can be read in at least two ways:
“where there is text it is only text and nothing more and this means that there is no person, no ‘thinking,’ no ‘feeling,’ no ‘sensations,’ no ‘world’”
Local reading: in the presented record, text is present; a person, thinking, feeling, sensation, and world are not parts of the text.
Global reading: no person, thinking, feelings, sensations, or world exists at all.
The first reading preserves the local full stop. The second is a separate universal claim and requires a separate warrant. The form itself does not establish whether either reading is uniquely warranted or intended.
When several readings remain possible, the text should not be silently corrected to fit one of them. The readings should be stated separately, and any unresolved ambiguity preserved until further context or support is available.
The full stop also blocks the substitution of a reader’s interpretation for the meaning of the supplied form.
Record, person, system
Record and event
The formula “a recording is a former sound” substitutes an event for a record. A more precise account distinguishes three stages:
a past acoustic event;
a record produced through transformation and storage;
a new acoustic event produced during playback.
These stages may be causally related. They are not identical. A record may preserve enough structure to be played back without becoming the past event.
Person
A message may contain the letter “I,” read as a pronoun, or the form “this annoys me.” Neither a person nor the sender’s state is part of the line. Reading the line as an expression of annoyance may be a well-supported attribution; it does not turn the attributed state into an observed fact. The absence of that state is not established either.
System
The forms “I think,” “I feel,” “I remember,” and a written step-by-step explanation are forms of textual output. They do not become thinking, feeling, memory, or a computational process. They do not establish the absence of those processes either.
Readable output is text. It does not become a process and does not establish the absence of a process.
Empirical work has shown cases in which written step-by-step explanations did not reflect causal influences on the answer [5]. This supports the need to distinguish explanatory text from an attributed process. It does not establish the practical effectiveness of the full-stop rule.
Scope
This document is itself text. Its typography, numbering, and references do not certify its truth, novelty, usefulness, or finality.
It does not claim that only text exists.
It does not claim that meanings, people, states, processes, or a world do not exist.
It does not claim either “I do not exist” or “I am everything.”
It claims neither an absolute separation of humans from nature nor the absence of all differences.
It does not claim that a claim expressed in text cannot be true, accurate, or well warranted.
It does not claim that all readings are equivalent.
It does not claim that a final interpretation-free capture level has been found.
Its claim is narrower:
On a selected textual surface, what a form names, the absence of what it names, a relation to it, and a selected reading must not be treated as parts of the text itself. The form of an assertion must not be treated as proof of its truth, accuracy, or status as knowledge. A boundary between written forms must not be treated as a boundary in their subject matter. Coordination between texts must not be treated as a direct record of understanding.
The basic distinctions themselves are not presented as discoveries. Existing work on signs, meaning, argument, form, and written explanations addresses related questions; references [1–5] are listed below. This document sets an earlier, local boundary and does not substitute that boundary for those theories. In particular, reference [4] takes a related position on form, meaning, and understanding. Its stronger conclusion about systems trained only on form is neither adopted nor rejected here.
The formulation alone does not establish the practical value of the full stop. To add value, the rule must change how an unsupported claim is handled without weakening claims that already state their inferential relation, and without resolving ambiguity on the author’s behalf. If it changes nothing, it is redundant.
Minimal operation
1. Copy the supplied text without correction.
2. Name the working capture level.
3. Place the full stop.
4. Check for four substitutions: (a) something named, its absence, a described relation, or a selected reading has been treated as part of the record; (b) the form of an assertion has been treated as proof of truth, accuracy, or knowledge; (c) a classification boundary has been projected into the subject matter; or (d) a coordinated response has been treated as a direct record of understanding.
5. If a claim about an author, state, understanding, event, process, source, world, truth, accuracy, or knowledge is needed, state it separately and identify its warrant.
6. If several readings remain possible, separate them instead of selecting one as the original meaning.
7. Keep the strongest local claim that does not treat later material as part of the earlier record.
There is text here. Full stop.
What is named does not enter with the name. Absence does not enter with negation. A relation does not enter with its description. A reading does not enter as the established meaning of the supplied form. The form of an assertion does not establish its truth, accuracy, or status as knowledge. A classification boundary does not become a division in the subject matter itself.
Text may be used to make claims about an author, state, event, process, source, referent, or world. Those claims may be true, well supported, and necessary. On this surface, however, they appear as later text; they do not become parts of the first text.
Nothing outside the text is inside the text. This does not mean that nothing exists outside the text. It means that the text does not decide that question.
Notes
1. Ferdinand de Saussure. Course in General Linguistics. 1916.↩
2. Charles S. Peirce. Collected Papers of Charles Sanders Peirce. 1931–1958.↩
3. Stephen E. Toulmin. The Uses of Argument. 1958.↩
4. Emily M. Bender and Alexander Koller. “Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data.” 2020.↩1 ↩2 ↩3
5. Miles Turpin, Julian Michael, Ethan Perez, and Samuel R. Bowman. “Language Models Don’t Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting.” 2023.↩1 ↩2