SKIP TO CONTENT
MINDLOOMbeta
RU
Mindloom Engine

Neurosymbolic Speech Analysis Engine

What goes in. What comes out. How the pipeline works. And honest positioning: what it is and what it is not.

Mindloom Engine is an experimental neurosymbolic system for classifying speech behavior. LLM understands context and nuance. Deterministic rules ensure reproducibility and transparency. Input: spontaneous speech. Output: a structural map—regimes, forms, vectors, frequency profile.
Pipeline

From text to structural map

Seven processing steps. Click for details.

01Input02Segmentation03Classification04SOL Vector05Aggregation06D-Index07OutputRegime (10)Layer (4)Form (171)Vector (5)Four classification dimensions
DEMO
Interactive engine demonstration
Enter text → get structural annotation in real time
Honest Position

What it is and what it is not

Mindloom is an original research program, not an established scientific discipline. Everything described here is not settled knowledge but a set of formalized hypotheses, tools, and preliminary results undergoing validation.

We are developing a research framework for structural analysis of speech behavior. The framework is formalized as a working ontological specification, partially implemented as an experimental engine, and is in the stage of internal validation with a plan for external verification.
WHAT MINDLOOM CLAIMS
Speech behavior contains computable structures — regimes, forms, layers — that can be formalized and detected automatically
These structures are not emotions, styles, or topics, but functional configurations of speech behavior
The specific set of 10 regimes, 4 layers, 171 forms is the current working version, not final truth
The proposed map generates testable predictions and enables seeing what cannot be distinguished without it
WHAT MINDLOOM DOES NOT CLAIM
Regimes "objectively exist like chemical elements." They are analytical categories — tools of distinction, not natural kinds
The ontology is "open." It is constructed. The question is not "is the map true" but "is it useful for prediction and description"
Accuracy scores prove "truthfulness" of categories. They show that the model reproduces annotations. This is necessary but not sufficient
Path to Validation

What is needed to move from framework to verified program

01Inter-annotator agreement (κ ≥ 0.6 for each SOL parameter)
02Convergent validity (Mindloom vs LIWC, BERT, RF Scale Fonagy)
03Predictive validity (RS predicts dialogue outcome)
04Preprint on PsyArXiv — opening for external criticism
05External replication — different researcher, different corpus
06Peer review — peer-reviewed publication
OBJECTIVE SELF-ASSESSMENT
6.7/10
20 criteria. Position is intentionally adjusted toward rigor. Current state is evaluated, not potential.
STRONG (8–9)
Distinctions, core clarity, epistemological honesty
MEDIUM (6–7)
Ontology, falsifiability, engineering, eval
DEFICIT (3–5)
IAA, convergent validity, predictive power
Mindloom is valuable insofar as it enables another person to see in speech what would otherwise remain undistinguished, and to do so with enough precision that it can be reproduced, verified, and applied.
READ NEXT
Ten Speech Regimes
One of the engine's principal classification targets — the ten speech regimes.
Protective Speech Architecture
What the engine is looking for in speech — the 13 levels of the protective cascade.
Systemic Hypotheses
What predictions the engine makes and how they could be falsified.
MINDLOOM COGNITIVE ARCHITECTURE LAB · 2026