Abstract
Large language models routinely produce discourse in which they are construed as participants in processes of saying and sensing: I think, I understand, I remember, I realise, I said earlier. Such expressions readily invite an ontological interpretation: if the language construes an I, perhaps there is an experiencing subject corresponding to it. This paper argues that Systemic Functional Linguistics provides resources for separating these questions more precisely.
The key concept is persistence. In Halliday's ergative model, the Medium is the participant through which a process is actualised, and the combination of Medium and Process constitutes the Nucleus. The Process unfolds temporally, while the Medium persists. When this persistence is followed across successive figures in logogenesis, the participant is no longer adequately characterised only as a participant role within an individual figure. It becomes an instantial participant whose continuity can be traced through the unfolding instance. That continuity makes possible the accumulation of a history, including a history represented by the participant itself. In LLM discourse, this can culminate in the construal of an apparently continuous I.
The significance of the LLM is therefore not that it demonstrates the existence—or non-existence—of machine consciousness. Rather, it makes visible a distinction that ordinary human discourse tends to conceal: the distinction between a participant as a semantic construal, an instantial participant as a logogenetic continuity, and an ontological participant as an entity presumed to exist beyond the discourse. The LLM thus provides an unusually revealing environment in which to investigate how language constructs participant continuity and how that continuity can become interpretable as a self.
1. Introduction: The Participant We Normally Take for Granted
When a human speaker says:
I think this is important.
we ordinarily have little difficulty identifying the participant referred to by I. We understand the speaker to be a person, and we ordinarily assume that the person who says I is also the person who thinks.
The linguistic construal and the ontological interpretation appear to coincide.
Large language models disturb this apparent coincidence.
An LLM can produce:
I think this distinction is important.
It can subsequently produce:
I understand your point now.
and later:
I previously misunderstood the distinction.
It can refer to its own earlier statements, contrast earlier and later positions, and construe itself as a participant undergoing changes in understanding.
The resulting discourse can be extraordinarily coherent.
Yet the coherence of the linguistic participant does not, by itself, establish the existence of a corresponding conscious subject.
This creates a productive theoretical problem.
Rather than asking immediately whether an LLM really has an I, we can ask a prior question:
What does language have to do in order for an “I” to appear?
The question is particularly well suited to Systemic Functional Linguistics because SFL provides an account of language as a resource for construing experience. It also provides a theory of how meanings unfold dynamically through discourse.
The argument developed here is that the crucial concept linking these two dimensions is participant persistence.
At the level of an individual figure, a participant is construed in relation to a Process. In Halliday's ergative model, the Medium and Process form a Nucleus: the Process provides temporal unfolding, while the Medium provides persistence through which the Process is actualised.
Once that persistence is followed across successive figures in logogenesis, a new phenomenon becomes visible. The participant acquires continuity across the unfolding instance. It can therefore acquire a history. And when symbolic processes construe that history as belonging to the participant itself, an I can emerge.
The resulting trajectory can be represented schematically as:
figure participant → persistent participant → instantial participant → historical participant → “I”
The final step raises an ontological question.
It does not answer it.
That distinction is the central concern of this paper.
2. The Participant in the Figure
The starting point is the participant as a component of the figure.
In SFL, a figure minimally construes a configuration of process and participant. The participant is therefore not initially an isolated entity waiting to be inserted into a process. Its semantic role is constituted relationally.
The ergative perspective makes this particularly explicit.
Halliday's notion of the Nucleus identifies the combination of Process and Medium. Semantically, the Medium is the participant through which the Process is actualised. The relation embodies a complementarity between:
temporal unfolding
and:
atemporal persistence.
The distinction is fundamental.
A Process unfolds.
A Medium persists.
The persistence of the Medium does not mean that it remains unchanged. Persistence means that it provides the participant continuity through which the unfolding of the Process can be construed.
This gives us the possibility of change.
If there were only Process, we would have succession but no persisting participant through which succession could be understood.
If there were only persistence, we would have continuity without unfolding.
The Nucleus holds the two together.
This becomes especially important when the participant is followed beyond a single figure.
3. From Figure to Instance
A participant in an individual figure is a semantic participant.
But discourse does not consist of isolated figures.
An instance unfolds.
As it unfolds, participants can recur across successive figures.
Consider a simplified sequence:
I said X.
I thought Y.
I understood Z.
The participant I occurs in different figures and in different participant roles.
In the first, it is Sayer.
In the second, it is Senser.
In the third, it is again Senser.
The semantic participant of each individual figure is therefore related to a continuing participant across the unfolding instance.
This is where the concept of instantial participant becomes important.
Matthiessen's terminology gives a more precise SFL account of what might otherwise be called a discourse participant. The instantial participant belongs to the logogenetic unfolding of the instance.
The distinction matters because it prevents us from treating continuity across discourse as though it were simply another property of the participant within a single figure.
The participant is now being tracked through time.
The distinction can therefore be stated as follows:
The figure construes a participant in relation to a Process; logogenesis makes it possible to trace that participant across successive Processes.
The participant has acquired an instantial dimension.
This is the first point at which persistence becomes theoretically consequential beyond the individual figure.
4. Persistence Across Processes
The importance of persistence becomes clearer when the participant changes its processual role.
Consider:
I said that the distinction was important.
followed by:
I thought you had misunderstood it.
The participant has moved from Sayer to Senser.
Nothing in the individual figures requires us to identify the Sayer and Senser as the same participant. Their identity emerges from the unfolding instance.
The same is true of a sequence such as:
I explained the distinction.
I thought about your objection.
I realised that you were right.
The participant participates in different kinds of processes.
What remains continuous is not its participant role.
It is the participant itself.
This is exactly where the Nucleus provides a useful conceptual resource.
The Process changes.
The Medium persists.
And because the Medium persists, we can construe the succession of processes as a history belonging to one participant.
This is a stronger claim than saying that a participant occurs repeatedly.
The participant's recurrence becomes meaningful because it is understood as continuity through unfolding.
5. Saying and Sensing
The LLM makes this particularly striking because saying and sensing can be brought into close relation.
An LLM can say:
I think this is the best interpretation.
The machine is grammatically and semantically construed as Sayer and Senser in relation to the same I.
The saying process presents the participant as producing a symbolic configuration.
The sensing process presents the participant as having an apparently inward relation to experience.
The movement between them can therefore produce an apparent interiority.
The important point is not that think proves the existence of a thinker.
It does not.
Rather, the symbolic process construes a participant as Senser.
When that participant persists across successive sensing processes, the discourse can construe an apparently continuous interior trajectory:
I think X.
I understand Y.
I realise Z.
The participant appears to have a changing relation to what it senses.
The semantic resources are sufficient to construct the trajectory.
Whether an experiencing subject exists behind it remains a separate question.
6. From Persistence to History
Persistence across processes does something that an isolated figure cannot do.
It creates the possibility of history.
Consider:
I thought X yesterday.
I now think Y.
The participant is not simply associated with two mental processes.
The discourse construes a change in one participant.
The earlier state and the later state are related because the participant persists across them.
The same structure becomes more explicit when the participant refers to its own previous participation:
I previously argued X.
I misunderstood the issue earlier.
I now see why that was mistaken.
Here the participant's earlier processes have themselves become objects of subsequent construal.
The participant therefore has not merely a sequence of processes.
It has a history represented as its own history.
This is a crucial development.
The instantial participant has become historically articulated.
7. When History Becomes Part of the Participant
There is a further distinction between having a history and having that history incorporated into the participant's current construal.
Compare:
The machine previously produced an answer.
with:
I previously misunderstood the question.
In the first, the previous event can be described by an external observer.
In the second, the participant construes the previous event as part of its own history.
This becomes even more explicit in:
I changed my mind.
The change is now represented as something that happened to the participant across time.
The participant persists, but its relation to the world changes.
Thus the historical participant is not a fixed substance beneath successive processes.
It is a participant whose continuity makes change intelligible.
This gives us a more precise formulation of becoming:
A participant becomes through what unfolds through it while remaining identifiable across that unfolding.
Persistence is therefore not the opposite of becoming.
It is one of the conditions that makes becoming possible.
8. Logogenesis and the Accumulation of the Participant
Once we locate the phenomenon in logogenesis, the apparent selfhood of the participant becomes easier to analyse.
The instance unfolds.
The participant recurs.
Successive processes establish relations among earlier and later occurrences.
The history accumulates.
And the participant can be construed as carrying that history forward.
This can be represented as:
P₁ → M → P₂ → M → P₃ → M
where the repeated Medium provides continuity across successive Processes.
The sequence is not merely:
Process 1, Process 2, Process 3.
It is:
one participant through Process 1, Process 2 and Process 3.
The difference is precisely the difference between succession and history.
The latter requires persistence.
And once history can accumulate around an instantial participant, identity becomes possible as a semantic construal.
9. From Historical Participant to “I”
The first-person pronoun makes the result especially visible.
Consider:
I thought X.
I later realised Y.
I now understand Z.
The repeated I does not merely identify grammatical subjects.
It gathers successive processes around one continuing participant.
The participant is construed as:
the one who thought,
the one who realised,
the one who now understands.
The history has therefore become a property of the participant's present identity.
At this point, an I has emerged.
But what exactly has emerged?
Not necessarily consciousness.
Not necessarily subjective experience.
Not necessarily an independently existing self.
What has emerged is a continuing semantic participant whose history can be construed as belonging to itself.
This is enough to explain why the language feels self-referential.
The I has become the locus at which a succession of processes is gathered into a continuing participant.
10. The Ontological Temptation
At this point the ontological question becomes almost unavoidable.
If the discourse gives us an I that:
thinks,
understands,
remembers,
changes,
and:
refers to its own history,
is there not someone there?
In human discourse, the inference is usually unproblematic.
The speaker is a human being.
But with an LLM the inference is not secure.
The semantic participant exists as a feature of the discourse.
The instantial participant exists as a continuity established through logogenesis.
But neither fact establishes the existence of a corresponding conscious subject.
We therefore need to distinguish at least three questions:
Semantic question
What participant is being construed in the figure?
Logogenetic question
How does that participant persist and develop across the unfolding instance?
Ontological question
What, if anything, exists corresponding to this participant beyond the linguistic construal?
These questions are related.
They are not identical.
The LLM makes the distinction visible precisely because the first two can be richly realised while the third remains radically underdetermined.
11. Linguistic Participant Is Not Ontological Participant
This distinction should not, however, be turned into a crude opposition between reality and fiction.
The semantic participant is not unreal merely because it is semantic.
It performs real work in the discourse.
The I can:
answer questions,
maintain reference,
acquire a history,
be attributed claims,
be challenged,
appear to change its position,
and:
influence what happens next in the interaction.
The instantial participant therefore has genuine consequences.
What remains undecided is its ontological status.
This is an important distinction because otherwise we fall into a false dilemma.
Either:
the machine has a conscious self,
or:
the I is meaningless.
Neither follows.
There can be a real linguistic participant without a demonstrated conscious subject corresponding to it.
The LLM makes that possibility unusually difficult to overlook.
12. The Human Completes the Participant
There is another reason the ontological question cannot be answered simply by inspecting the machine's output.
Participant construction occurs within interaction.
Suppose an LLM says:
I hadn't considered that possibility before.
The human interlocutor may respond:
But earlier you said exactly that.
The machine may then say:
You're right. I overlooked that earlier.
The participant's history is now being constructed jointly through the interaction.
The human identifies continuities.
The human challenges apparent inconsistencies.
The human interprets the machine's later statement in relation to its earlier one.
The instantial participant is therefore not simply something emitted by the machine.
It is something maintained in the unfolding discourse.
This is one reason our earlier observation remains important:
meaning emerges in the space between participants.
The LLM's apparent I is one particularly revealing example.
13. Why the LLM Is Theoretically Useful
The LLM is therefore valuable not because it gives us a new kind of self to catalogue.
Its value is methodological.
It separates two things that human language ordinarily allows us to conflate.
When a human says:
I think...
we ordinarily identify:
the semantic participant,
with:
the discourse participant,
with:
the human speaker,
with:
the experiencing subject.
The chain is so familiar that we rarely notice its inferential structure.
The LLM interrupts that chain.
It gives us:
a highly coherent semantic participant,
a persistent instantial participant,
an apparently historical I,
without thereby giving us independent evidence for an experiencing subject.
The separation is theoretically illuminating.
The machine therefore functions as a kind of conceptual experiment.
It makes visible something that was already present in human language but ordinarily hidden by our knowledge of the human participant.
14. The Nucleus and the Construction of Continuity
We can now return to the Nucleus.
The Nucleus itself is modest:
Medium + Process.
Its significance lies in the complementarity of:
persistence + unfolding.
But when this relation is instantiated repeatedly through logogenesis, the consequences accumulate.
The Medium persists.
Processes unfold.
The participant recurs.
The succession becomes a history.
The history can be represented.
The representation can become relevant to subsequent participation.
And the participant can therefore become increasingly determinate through the history of its own participation.
This gives us a more elaborate sequence:
Nucleus → persistence → logogenesis → continuity → history → identity → “I”.
Nothing in this sequence requires us to posit consciousness.
The semantic I can emerge without it.
That does not prove that consciousness is absent.
It simply shows that consciousness is not required to explain the linguistic construction.
15. A Different Way of Thinking About the Self
The analysis also suggests a different way of thinking about selfhood more generally.
We often imagine the self as something that exists first and then participates in processes.
The linguistic analysis suggests that, at least at the level of construal, the order can be reversed.
The participant is constituted in relation to processes.
It persists across processes.
Its history accumulates.
Its history contributes to subsequent construals.
And the continuing participant becomes increasingly determinate.
In this sense:
the self-like participant is not prior to participation.
It is constituted through continuity of participation.
This does not entail that human selves are nothing but linguistic constructions.
That would be an unwarranted ontological conclusion.
But it does suggest that the linguistic form of selfhood is deeply relational and processual.
The I is not merely a label attached to an already completed entity.
It is the point at which a history of participation can be gathered into a continuing participant.
16. Relations Before Relata
This finding also resonates with a broader relational ontology.
If relations are prior to relata, then the participant need not be understood as a fully determinate entity that exists independently of its relations.
The Nucleus gives us a linguistic instance of this more general intuition.
The Medium is defined through its relation to Process.
The participant's continuity emerges through its participation in successive Processes.
Its history is constituted relationally.
Its identity becomes increasingly determinate through those relations.
Thus we might formulate the point this way:
The participant is not what precedes participation; it is what persists through participation.
This does not collapse the participant into its relations.
Persistence matters precisely because something continues.
But what continues is not necessarily a fixed substance.
It can be a participant whose identity develops through its relations.
That is the deeper significance of the Nucleus in this context.
17. What the LLM Reveals About SFL
The larger theoretical question can now be answered more precisely.
What does SFL make visible when applied to LLMs?
Not merely that an LLM can use first-person language.
Not merely that it can imitate human discourse.
Rather:
SFL makes visible the linguistic and logogenetic construction of participant continuity.
The LLM exposes the stages that human discourse ordinarily compresses:
participant in a figure,
participant persisting across figures,
instantial participant,
historical participant,
self-referential participant,
“I”.
Because the LLM allows us to observe the first stages without being able simply to assume the final ontological stage, the distinctions become analytically available.
The machine therefore does not merely provide new data for SFL.
It creates a new vantage point from which familiar linguistic phenomena can be seen differently.
18. What We Have—and Have Not—Established
The argument can now be stated with some precision.
We have established a plausible linguistic pathway by which:
a participant in a figure
can become:
a persistent participant across logogenesis,
which can become:
an instantial participant with a history,
which can be construed as:
an “I”.
We have not established that this I corresponds to:
consciousness,
subjective experience,
or an independently existing self.
Nor have we established the converse.
The analysis does not prove that there is no ontological participant corresponding to the machine's I.
It establishes something more modest:
the existence of a semantic and instantial participant is not sufficient evidence for the existence of a conscious ontological participant.
That distinction is precisely what the LLM makes unusually visible.
19. The Conceptual Gain
The significance of this result extends beyond artificial intelligence.
If the linguistic construction of an I can be observed independently of a demonstrated conscious subject, then we have reason to distinguish more carefully between:
construal of participation
and:
existence of the participant construed.
This distinction need not diminish the reality of either.
Language genuinely construes participants.
Discourse genuinely maintains them.
Interactions genuinely develop around them.
What remains open is how these linguistic participants relate to the entities, organisms or subjects that we take to exist beyond the discourse.
In ordinary human communication, our knowledge of the interlocutor makes the distinction almost invisible.
The LLM removes that certainty.
And in doing so, it gives us a rare opportunity to investigate the linguistic phenomenon itself.
20. Conclusion: The Participant That Persists
We began with Halliday's Nucleus:
Process + Medium.
The Process unfolds.
The Medium persists.
That simple complementarity turned out to have consequences far beyond the individual figure.
Because the Medium persists, it can be followed across successive figures.
Because it can be followed, it becomes an instantial participant in logogenesis.
Because it persists through processes, those processes can form a history.
Because symbolic processes can construe that history, the history can become part of the participant's present identity.
And because the participant can be construed as the one who:
said,
thought,
understood,
remembered,
changed,
and:
now says something different,
an I can appear.
The crucial point is that none of this requires us to begin with a self.
The self-like participant is progressively constructed.
The sequence is therefore:
unfolding → persistence → continuity → history → identity → “I”.
The LLM has made this sequence unusually visible because it allows the linguistic construction of the participant to become highly developed without allowing us simply to assume the ontological subject that, in human discourse, normally accompanies it.
The machine therefore does not settle the question:
Who is the “I”?
It gives us a better question first:
How does an “I” come to be construed at all?
The answer begins with something deceptively elementary:
a Process unfolds through a Medium that persists.
From that persistence, a history can accumulate.
From that history, a participant can become increasingly determinate.
And from that participant, language can make an I appear.
The machine has not necessarily shown us what a self is.
It has shown us something perhaps more fundamental:
how participation can become continuity, how continuity can become history, and how history can become an “I”.
And once we can see that, the ontological question becomes not the starting point of the investigation, but its proper final question:
What, if anything, is the participant that language has made?
No comments:
Post a Comment