At this point in the argument, a familiar counter-move appears.
If meaning cannot be reduced to probability alone, perhaps it can be recovered by adding context. Perhaps once we situate probabilistic patterning within rich situational descriptions—participants, activities, purposes, settings—the gap between fluency and meaning will finally close.
This move feels promising because context does matter. Meaning is always made in situations, never in a vacuum. But this promise is illusory.
Context conditions meaning; it does not determine it. And no enrichment of situational description can convert probabilistic patterning into construal.
1. Why Context Looks Like the Missing Ingredient
The appeal to context usually arises from a correct diagnosis paired with an incorrect remedy.
The diagnosis: abstract symbol statistics are insufficient. Meaning varies with situation, with social relation, with what is being done. Strip language of context and you strip it of life.
The remedy then proposed: add context back in—metadata, situational variables, embeddings of use—and meaning will emerge.
But this treats context as though it were a supplementary input to an otherwise complete mechanism. It assumes that once enough situational parameters are specified, meaning becomes computable.
This assumption mistakes conditioning for determination.
2. Context as Semiotic Environment, Not Causal Engine
Within a Hallidayan framework, context is not a bundle of causes that produce meanings. It is a semiotic environment within which meanings are possible.
Field, tenor, and mode do not determine what is meant. They shape the space of relevant options. They weight potential. They make some construals more expectable and others less so.
But expectancy is not causation.
Even in highly routinised situations, meaning is not forced. A situation constrains without compelling. Speakers and writers still make perspectival cuts; listeners and readers still construe phenomena.
Context, therefore, belongs with system rather than instance. It structures potential; it does not execute actualisation.
3. Conditioning Without Determination
This distinction—between conditioning and determination—is easy to state and difficult to hold.
To say that context conditions meaning is to say that it:
shapes the semiotic resources likely to be deployed
biases selections within systems of potential
stabilises expectations across social coordination
To say that context determines meaning would be to claim that, given sufficient situational description, meaning follows as an outcome.
The latter is false.
Meaning is not an output of context. It is an act within context.
No situational description, however detailed, eliminates the need for construal. It merely frames it.
4. Situation Does Not Rescue Probability
This is where probabilistic reduction quietly re-enters under a different name.
Once context is treated as a set of variables, it becomes tempting to fold it into the probability space: probabilities conditioned on context. Given situation S, the likelihood of expression E increases. With enough data, the model approximates situational sensitivity.
But this manoeuvre does not cross the ontological boundary identified earlier.
Probabilities conditioned on context are still probabilities over second-order residues. They describe how construals have tended to cluster under similar conditions. They do not produce a construal.
Adding context refines the grain. It does not change the level.
5. Why LLMs Feel Context-Sensitive
Large Language Models are increasingly praised for their contextual awareness. They track discourse history, adapt to prompts, and respond differently across apparent situations.
This is real, but it is not what it seems.
What LLMs model is not context as lived situation, but context as textually recoverable pattern. They infer situational cues from linguistic traces and adjust probabilities accordingly.
This produces impressive alignment. It does not produce situated meaning.
The model’s sensitivity is to correlations among residues, not to situations as experienced.
6. The Persistence of the Error
Why, then, does the idea persist that context might rescue probabilistic accounts of meaning?
Because context is where meaning feels anchored. It is where language meets life. And so it is tempting to think that once context is formalised, meaning will follow.
But context is not a substitute for construal. It presupposes it.
Situations do not mean. People mean in situations.
7. The Cut Reaffirmed
We can now restate the asymmetry one final time:
Probability describes patterned residues of meaning-making.
Context conditions which patterns are likely.
Meaning arises only through construal within situations.
No accumulation of probabilistic sensitivity, no enrichment of situational variables, dissolves this structure.
In the next post, we will turn to agency, intention, and responsibility—not to humanise meaning, but to show why meaning is irreducibly an act, not an outcome.
Context matters.
But it does not decide.
The cut still holds.
No comments:
Post a Comment