Skip to content

v2-io/hedge-embeddings

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

90 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Experiments in Epistemic Hedging Geometries

...

Bayesian Degree of Belief

The standard view: "Degree of belief" is a philosophical interpretation of the probability calculus — a choice you make about what the formalism means.

What this data suggests: Degree of belief is a measurable geometric property of natural language that exists independently of any philosophical framework. Hedge words occupy positions on a calibrated linear axis in embedding space — not because anyone trained the models on Mosteller data, but because that's how these words function in billions of sentences of actual language use.

The classic objection to Bayesianism — "whose beliefs?" — gets a concrete answer: the language community's beliefs, as they're encoded in distributional structure and recoverable via projection. The consistency across 5 architecturally diverse models, independent psychometric data, and (potentially) LLM explicit reasoning would constitute convergent measurement from radically different methods — the hallmark of a real phenomenon.

The corpus analysis would close the causal loop (what distributional patterns create the geometry?), and the LLM decomposition would add a third measurement class (explicit reasoning about probability, as opposed to implicit geometric encoding or elicited human judgment).

Whether this is a section in the current paper, a companion paper, or the seed of a broader research program depends on how deep you want to go. The empirical paper is publishable now. The philosophical argument could make it memorable.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors