Deixis: the words a chat export cannot resolve
Deixis is the class of words whose reference comes from the situation of speaking, not from the word. A chat export keeps every one of them and loses what each one pointed at.
Deixis is the class of expressions whose reference is fixed by the situation in which they are used rather than carried in the word itself: here, now, this, tomorrow, that one, I, you. A chat export preserves those words exactly and discards the situation that fixed what they referred to, which makes them the words in a transcript that look readable and are not.
Read back an old conversation and most of it still works. The lines that stop working are the short ones: on the way, this one, same as yesterday, tell her tomorrow. That class of words has a name, an origin and a literature, and knowing which words belong to it is what separates reading a transcript from misreading it without noticing.
The words that point at something outside the file
In philosophy of language the term is indexical. The Stanford Encyclopedia of Philosophy entry on indexicals defines one as, roughly speaking, a linguistic expression whose reference can shift from context to context, and gives I, here, today, yesterday, he, she and that as paradigmatic examples. It notes that indexicals are commonly called context-sensitive expressions.
Linguistics covers the same ground under the name deixis and calls the words deictic expressions. The two traditions meet in the entry itself, which records that the indexical uses of pronouns such as he and she are sometimes called deictic or demonstrative uses, attributing the point to Kaplan 1989a and Partee 1989.
The entry’s own illustration is the whole problem in one line. It gives the sentence “He was born in Detroit” followed by the bracketed note “[Pointing at John]”, and says pronouns used that way would typically appear in utterances accompanied by pointing gestures. Write the sentence down and the sentence survives. The pointing does not.
Most words in a chat are not like that. Invoice, Marseille and the colour green mean the same on any screen on any day. A deictic word carries an instruction instead: find who is speaking, find when, find the thing being indicated, then substitute. Export the conversation and you keep the instruction and lose every input it was written to run on.
One unvarying character, a content that changes every time
The sharpest version of the distinction is David Kaplan’s. The Stanford Encyclopedia of Philosophy entry on indexicals records that many philosophers hold indexicals to have two sorts of meaning: linguistic meaning, or character, a term it attributes to Kaplan 1989a, and content. In the entry’s words, every indexical has a single unvarying character but may vary in content from context to context. It names Kaplan’s work Demonstratives as the source of most of the list it works from.
That is the shape of what an export does to a conversation. The character is the rule, it is carried by the word, and it is copied into the file perfectly. The content is what the rule produced on the day, and it is not in the file at all. The entry’s own example of the gap: when both John and Mary utter “I am hungry”, Mary says that she is hungry and John says that he is hungry. One character, two contents, from an identical string.
A transcript, read this way, is a file of characters with the contents stripped out. Nothing in it is corrupted. The rules are intact and there is nothing left to apply them to.
Why this and that need a gesture or a shared background
These words do not all fail the same way, and the Internet Encyclopedia of Philosophy entry on demonstratives and indexicals draws the line that matters. It separates pure indexicals, whose content is fixed automatically in a context in virtue of their meaning, with I, today and actually as its examples, from true demonstratives such as this and that, which it says require some kind of additional supplementation in a context in order to refer successfully. Both terms, it notes, are due to Kaplan. A true demonstrative, in its wording, requires something of the speaker, some kind of gesture or some kind of special intention, in order to resolve what the speaker is referring to.
Read against a chat file, that is a hierarchy of damage. I still has a rule a reader can run, because the message has a sender. Yesterday still has one, if the timestamp survived the export. This one has no rule left to run, because what it needed was never a word in the first place.
The same entry gives the term’s origin: indexical is due originally to Charles Sanders Peirce, who introduced it as part of a threefold theory of signs. It summarises Peirce on this and that as a condition on the audience. The audience of a successful use of a demonstrative can infer the existence of the object referred to, and if the audience cannot infer it, the use has not been successful. It adds that what a speaker refers to with this depends on a further feature of the context: either the speaker makes some gesture, or there is enough common knowledge in the background for the audience to identify the object intended.
A chat export gives a reader neither. The gesture was never recorded and the shared background was never typed, so on that condition many demonstratives in a transcript are uses that cannot succeed for a later reader.
Fillmore’s 1971 lectures anchored deixis in who, where and when
The linguistics side has a datable origin. CSLI Publications at Stanford publishes Lectures on Deixis by Charles Fillmore, and its page for the book states that the lectures reprinted there were given in Santa Cruz in the summer of 1971. The publisher’s description, a secondary statement of the position rather than Fillmore’s own words, is also the most useful definition for anybody reading a transcript. Every language, it says, has lexical items and grammatical forms that can be interpreted only when the sentences containing them are understood as anchored in some social context, and that context must identify the participants in the communication act, their location in space, and the time during which the act is performed.
Three anchors, then: who, where, when. Each survives an export differently. Participants survive in part, as display names or masked placeholders, which is a label on a message rather than a person in a room. Location is almost never written down, because nothing in a chat file records where either person was sitting, which makes here and over there the irrecoverable cases. Time survives best, as a timestamp on every line.
The same page records that Fillmore goes on to discourse deixis and social deixis, describing the second as the aspect of sentences that reflect or establish or are determined by realities of the social situation in which the speech act occurs. A written record holds that category reasonably well, and it overlaps with speech act theory read against text messages.
Tomorrow in an old message resolves against the reader
This kind of failure does not announce itself, which is the reason the category is worth naming. A corrupted file does. “See you tomorrow” does not: the word is spelled correctly, it has a meaning anyone knows, and nothing in the file marks it as the point where the reading stopped being determinate.
Philosophy has argued directly about the written case. The Stanford Encyclopedia of Philosophy entry on indexicals discusses sentences inscribed and read later, and records a proposal from Michaelson 2014 that indexicals have characters shifting with the medium of communication, so that now refers to the time of speaking in a live conversation but to the time of reading when written. The entry does not accept it, arguing the medium is too coarse grained for the job, because a postcard and a will point to different referents although both are written. It offers, as an adaptation of a case from Bianchi 2014, a note left for a butler reading “I am leaving you in charge of my affairs”; when the writer’s wife reads the note, the entry says it seems plausible that you refers to the butler and not to whoever reads it.
Whichever analysis is right, the disagreement is itself the useful finding. Specialists have not settled what a written you or now refers to once the reader is not the person it was written for, and a transcript reader is in that position on every line.
This is a narrower question than whether two people understood each other at the time, which is what grounding in communication is about, and narrower than whether a misunderstanding got fixed, which is conversational repair. Deixis is mechanical by comparison: a reference was computed from the situation, and the situation is gone.
The referent Portraitor never receives
In one case the missing referent is not merely absent from the file but excluded from the reading by design, which is worth stating precisely because it is checkable on a published page.
Portraitor’s privacy policy states that chat exports often contain photographs, videos and voice recordings, that the Service never opens them and never transmits them, and that only the written messages are read, on your device, with only their masked text sent. So where a message said “look at this” and the thing indicated was the image attached beside it, that referent is not degraded in what Portraitor reads. It is not in the input. The same policy states that before any part of a conversation is transmitted, the browser removes names, email addresses, telephone numbers, postal addresses and credentials and replaces them with anonymous placeholders, entirely on your own device, and that this removes who is speaking without removing what was discussed.
The attachments clause is set out in what happens to the photos in a chat export and the identifier clause in does a chat analysis remove names. Both report the same page.
The consequence for deixis is worth keeping narrow. Any tool reading a chat export is reading deictic words whose referents are missing, which is a property of exports rather than of one tool. For the subclass whose referents were attachments or personal names, Portraitor’s published position is that those referents were never part of what it received. That is a fact about the architecture rather than a quality of the reading, and one reason a Portraitor portrait describes how a person writes rather than what they were pointing at.
Where reading an export this way runs out
Naming deixis tells you which words to distrust. It recovers no referents, and little else follows from it.
It cannot tell you what this was: if the thing was never typed, no amount of reading produces it, and a confident guess is still a guess. It does not make a count of deictic words into a measure of anything about a person, and nothing cited here supports reading a frequency as a trait. It says nothing about whether either person understood the other at the time, because that needs the turns that followed, and those are sometimes not in the export either. Where a tool does the reading, every one of those limits applies to the tool, and the general version of that argument is in what a chat analysis cannot tell you about someone.
Frequently asked questions
Frequently asked questions
What is deixis in simple terms?
Deixis is the use of words whose reference is worked out from the situation of speaking rather than carried in the word: here, now, this, tomorrow, I, you. The Stanford Encyclopedia of Philosophy entry on indexicals defines the parallel philosophical term, an indexical, as roughly a linguistic expression whose reference can shift from context to context.
What is the difference between deixis and an indexical?
They cover close to the same ground from two traditions, deixis in linguistics and indexical in philosophy of language. The Stanford Encyclopedia entry records that the indexical uses of pronouns such as he and she are sometimes called deictic or demonstrative uses, attributing that point to Kaplan 1989a and Partee 1989.
Who came up with deixis and indexicals?
Three attributions, each from a page that states it. The Internet Encyclopedia of Philosophy says the term indexical is due originally to Charles Sanders Peirce. The Stanford Encyclopedia says character, the name for an indexical’s single unvarying linguistic meaning, is due to David Kaplan, 1989a, and names his work Demonstratives. CSLI Publications at Stanford, publishing Lectures on Deixis by Charles Fillmore, says the lectures reprinted there were given in Santa Cruz in the summer of 1971.
Which deictic words survive a chat export and which do not?
Time survives best, person partly, place almost not at all. Timestamps usually come through, so yesterday and last week can often be resolved; participants come through as display names or masked placeholders; and nothing in a chat file records where either person was, so here and over there have nothing to resolve against. The publisher’s page for Fillmore’s Lectures on Deixis names those same three anchors as what a deictic form must be anchored in.
Can a chat analysis work out what this referred to?
No, not when the referent was never written down. The text is all there is, and a demonstrative that pointed at an object in the room left nothing in the text to recover. Portraitor’s privacy policy adds a second reason in its own case: it states that the Service never opens and never transmits the photographs, videos and voice recordings an export contains, so an attachment that a this pointed at is not part of what gets read.
The deictic words are all still in the file. What they pointed at never was.
Your conversations, read the way you would read someone else’s.
Portraitor takes a chat export you already have, anonymises it in your browser, and returns a structured reading of how you come across. Identifying details are removed in your browser before anything is sent.
See how it works →- Braun, David and Alexandru Radulescu. Indexicals. Stanford Encyclopedia of Philosophy, substantive revision 3 September 2026.
- Georgi, Geoff. Demonstratives and Indexicals. Internet Encyclopedia of Philosophy.
- Fillmore, Charles J. Lectures on Deixis. CSLI Publications, Stanford University. Publisher page for the reprint of lectures given in Santa Cruz in 1971.
- Portraitor, Privacy Policy, version 1.3, section 2.