The Empirical Linguistic Engine (ELE)

A Unified Architectural Framework Across the P→P→L→C→C Continuum

(Physics → Physiology → Linguistics → Cognition → Communication)

---

I. Introduction: The Mandate for an Empirical Linguistic Engine (ELE)

The development of the Empirical Linguistic Engine (ELE) requires a decisive move away from purely abstract linguistic formalism toward a unified, recursive, and empirically validated computational architecture.

This architecture must:

• Integrate physical and biological constraints
• Respect cognitive and social theories
• Define explicit interfaces, measurable units, and operational principles
• Honor the causal hierarchy:

Physics → Physiology → Linguistics → Cognition → Communication
(P→P→L→C→C)

The Necessity of Grounding and Unification

Historically, the study of language has been fractured:

• Generative linguistics (Chomsky, 1950s) posits an innate Language Acquisition Device (LAD), a specialized, modular system devoted exclusively to language. [1]
• Cognitive linguistics (1970s onward) rejects strict modularity, arguing that language and cognition share mechanisms, and that language is fundamentally embodied and situated. [3]

From this perspective, traditional analyses focused solely on phonetics, syntax, and semantics are insufficient, especially when explaining communication deficits in everyday social interaction. [5]

The ELE adopts an integrative stance:

• It leverages the computational precision of structured representations.
• It mandates adherence to physical and physiological grounding.
• It directly addresses critiques of large language models (LLMs) that claim text-only training is ungrounded and lacks “real understanding”. [6]

ELE does this by explicitly integrating empirical constraints from the physical world into its architecture.

The Computational Imperative: Recursion and Multi-Scale Architectures

Human language is:

• Hierarchical
• Productivity-driven
• Capable of generating infinite expressions from finite means

The key mechanism enabling this is recursion—the embedding of structures within themselves. [8]

This functional requirement induces architectural constraints for any computational linguistic engine.

Modern neural network research shows:

• Complex language systems spontaneously develop hierarchical temporal structures that align with linguistic levels, without explicit pre-programming. [9]
• Multi-Timescale Recurrent Neural Networks (MTRNNs) self-organize into distinct timescales:
• Short timescales (~0.17 words): phonological processes
• Medium timescales (1–10 words): morphological and syntactic structures
• Long timescales (≈360,000+ words): lexical and semantic relationships [9]

This validates a functional separation of linguistic strata and supports the adoption of Recursive Linguistic Models (RLMs). [10]

RLMs:

• Depart from flat, purely sequential processing
• Allow recursive querying of:
• the model itself, or
• an external environment (e.g., a REPL)

This is crucial for:

• Managing vast context lengths
• Modeling human-like discourse
• Handling situated pragmatics at scale

---

II. Stage 1: Physics — The Aerodynamic Foundation

The linguistic chain begins in physics, not in abstract grammar.

At the foundational level:

• Air pressure and airflow are manipulated.
• These set material limitations and temporal boundaries for language.

Acoustic and Aerodynamic Principles

Speech production (phonation) depends on:

• Vital Capacity (VC) — maximum exhaled air volume after maximal inhalation
• VC and derived metrics (e.g., Phonation Quotient, Estimated Mean Airflow Rate) inform:
• vocal stamina
• maximum utterance length [12]

Key control variable:

• Lung (subglottal) pressure → drives self-oscillation models of the vocal folds. [14]

Outputs at this stage:

• Phonation Trigger
• Acoustic waveform

Examples:

• VC (mL) defines maximum temporal bounds of continuous utterance
• Typical norms:
  • Males ≤ 39 years: ~3530 mL
  • Females ≤ 39 years: ~3080 mL [12]

These volumes constrain Maximum Phonation Time, which in turn:

• Sets natural boundaries for breath groups and prosodic phrases.
• Tethers abstract linguistic units like “phrase” to physical respiratory capacity. [15]

A realistic ELE must therefore:

• Initialize speech generation with real-time or modeled aerodynamic data.

Historical-Conceptual Grounding: Pneuma

The linkage between air and cognition is ancient:

• Greek Pneuma (πνεῦμα) = “breath”, “wind”, “spirit”. [16]
• In classical thought, pneuma was seen as:
• inhaled air transformed into a vital spirit
• traveling through organs to the brain
• distributed via nerves as “animal spirit” [18]

Anaximenes (~500 BC) equated:

• Soul (psyche) = air (aer), breath (pneuma), and world-encompassing medium. [16]

Though biologically outdated, this framework anticipates ELE’s causal demand:

External air → physiological flow → cognitive control

---

III. Stage 2: Physiology — Embodied Processor and Motor Command

Here, bulk aerodynamic input becomes:

• Precisely controlled
• Temporally sequenced
• Biomechanically embodied motor output

Mechanisms of Phonation and Muscle Control

Controlled airflow is converted into oscillation via the larynx, regulated by:

• Fine-grained laryngeal muscle control
• Low-dimensional self-oscillation models capturing:
• body-cover differentiation of the vocal folds
• primary vibration modes (shear and compressional) [14]

Neural programming at this stage:

• Converts normalized activation levels of key muscles:
• Cricothyroid (CT)
• Thyroarytenoid (TA)
• Lateral Cricoarytenoid (LCA)
• Posterior Cricoarytenoid (PCA)
• Into physical quantities:
• vocal fold strain
• adduction
• glottal convergence
• stiffness [14]

Critical output:

• Pneuma (Motor Command) — the patterned neural command controlling glottal aerodynamics.

This is the modern physiological analog of the ancient pneuma.

Neurophysiological Constraints and Embodiment

Physiology is the causal bridge between:

• Abstract linguistic form
• Concrete motor action

Sequential motor control:

• Enforces precise timing
• Shapes articulation
• Suggests non-arbitrary foundations for:
• phonotactic rules
• linear order in syntax [19]

Evidence for embodied cognition:

• Language may rely on a “language prewired brain” built upon the Mirror Neuron System (MNS). [3]
• Language processes display interaction-dominant dynamics:
• heavily dependent on situational variables at fine timescales
• word recognition and lexical mapping depend on object visibility and actionfulness in context [20]

Thus:

Proper analysis of language must be conducted at the level of organisms + environment. [20]

Architecturally, ELE must:

• Be non-sequential only (not a strict pipeline)
• Be interaction-dominant and recursive at the sensory input layer
• Dynamically link:
• physiological and initial linguistic layers
• with an evolving external environment model from later cognitive stages

---

IV. Stage 3: Linguistics — Formal Hierarchical Structure (L-Units)

This stage:

• Formalizes structured organization of the linguistic signal
• Transforms physiological outputs into symbolic representation

Language’s power lies in its hierarchical structure, enabling infinite productivity. [8]

Classical Stratification and Hierarchy

Standard hierarchy:

• Phonology
• Studies sound systems
• Basic unit: Phoneme — smallest sound unit that can distinguish meaning (/p/ vs /b/) [8]
• Morphology
• Studies internal structure of words
• Basic unit: Morpheme — smallest meaningful unit (e.g., plural “-s”) [8]
• Syntax
• Rules for combining words into larger constituents
• Basic objects: phrases and sentences
• Key property: Recursion — embedding phrases within phrases [8]

Generative models (e.g., Government and Binding Theory):

• Propose “movement” and traces (silent copies) in sentence structure
• Psycholinguistic evidence supports traces as part of mental representation. [22]

Defining the Atomic Units of Meaning: The Sememe

Semantics sometimes requires a finer granularity than morphemes.

The Sememe (or seme):

• An indivisible, atomic unit of meaning [21]
• Historically: a “definite idea-content expressed in some linguistic form” [23]

Example:

• “Triangle” = sememe “three-sided straight-lined figure” [23]

Computational Hierarchy and Embodiment

Empirical evidence:

• ERP/MEG studies show morphologically complex words elicit larger responses than monomorphemic words — supporting active morphological decomposition during access. [24]

Neural architectures mirror this:

• Transformer models (e.g., BERT) exhibit systematic info distribution:
• Early layers (1–3): phonological/morphological segmentation
• Middle layers: syntactic structure
• Late layers: lexical and semantic relationships [9]

This:

• Empirically validates functional separation of linguistic levels.
• Confirms that Sememes (atomic meaning units) must be:

Grounded through sensorimotor simulations (Stage 2),
linking semantics with embodied cognition and action. [15]

Thus, there is a biologically plausible feedback loop between:

• Linguistics (Stage 3)
• Physiology (Stage 2)

---

V. Stage 4: Cognition — Mental Representation and Architectures

Cognition:

• Acquires, stores, and applies linguistic knowledge
• Interfaces formal language structure with the external world

Language–Cognition Interaction

Language and cognition:

• Interact closely rather than being strictly identical or fully separate. [3]

Cognition:

• Builds internal models of the world

Language:

• Acts as a storehouse of cultural wisdom
• Serves as a teacher, adapting cultural knowledge to concrete life situations. [3]

Psycholinguistic Mechanisms: Storage, Retrieval, Generalization

Language processing hinges on:

• Sensory, short-term, and long-term memory coordination. [26]
• Chunking: grouping elements into manageable units

Short-term memory:

• Holds roughly 5–9 chunks [26]
• Directly shapes parsing efficiency and lexical retrieval

Metarules:

• Capture generalizations about language structure
• Do not list structures themselves
• Instead define classes of permissible structures [27]

In frameworks like Generalized Phrase Structure Grammar (GPSG):

• Metarules, Feature Co-occurrence Restrictions, and the Head Feature Convention:
• compactly encode structural generalizations [27]

Cognition also governs:

• Conceptual metaphor
• Prototype categorization [29]
• Figurative language processing (metaphors, idioms) [28]
• Mnemonics and associative networks for long-term recall [30]

Computational Architectures and Limits

To model long-term dependencies and context:

• ELE must use Recursive Linguistic Models (RLMs) [10]
• RLMs:
• handle vast context via recursive subqueries
• mimic attention and memory mechanisms across long discourse

Caution:

• fMRI-based semantic mapping (brain regions vs. conceptual fields) suffers:
• signal loss in temporal regions
• performance-related variability [32]

Thus, ELE’s cognitive validation demands:

• Controlled performance variables
• Multi-modal methods (EEG/MEG, behavioral data, etc.)

---

VI. Stage 5: Communication — Situated Pragmatics and Intentionality

This stage deals with actual use of language in:

• Social contexts
• Environmental situations

Here, meaning is defined not only by structure but by:

• Intention
• Context
• Social norms

Situated Context and Social Layers

Pragmatics:

• Studies how utterances communicate beyond literal meaning and grammar [34][35]
• Focuses on:
• implications
• inferences
• attitudes
• situational context

Communication operates across social layers [28]:

• Social Norms layer:
• unwritten rules
• respect, politeness, body language
• when/how to speak
• Friendships layer:
• humor, empathy, in-jokes
• slang and non-literal uses

Defining the Unit of Interaction: The Pragmeme

Analogous to Sememe (semantic atom), Pragmeme:

• Proposed to clarify “indirect speech act” phenomena [36]
• Defined as:
• a unit realized through pragmatic acts
• tightly bound to situation and context [36]

A Pragmeme may:

• Arise through Pragmaticalization
• Carry its own illocutionary force [37]

Measuring it requires:

• Rich pragmatic assessment protocols
• Inclusion of:
• linguistic
• extra-linguistic
• paralinguistic signals [5]

Cross-cultural work needs coding schemes that distinguish:

• Core Code Categories (replicable, e.g., syntactic downgrading)
• Situated Code Categories (culture-specific, context-bound) [38]

Intentionality and Theory of Mind

Successful communication depends on:

• Inferring others’ beliefs, intentions, and perspectives
• This is Theory of Mind (ToM) [39]

For ELE:

• ToM-like inference is required for full pragmatic competence
• Recent work shows advanced LLMs can pass false-belief tasks, suggesting:
• sufficient structural complexity to model basic ToM reasoning [39]

The Pragmeme:

• Is the highest-order grounding unit:
• Validates and adapts cognitive constructs (Stage 4)
• Constrains them via real-world environment and social dynamics [20][36]

This demands an interaction-dominant architecture:

• Continuous adaptation
• Dynamic sensitivity to context and norms

---

VII. The Empirical Linguistic Engine Architecture and Conclusions

The ELE unifies five domains:

1. Physics
2. Physiology
3. Linguistics
4. Cognition
5. Communication

In this framework:

• Physical constraints start the chain
• Physiology embodies it
• Linguistics structures it
• Cognition indexes and recursively manipulates it
• Communication situates and validates it

The Integrated P→P→L→C→C Model

The architecture must:

• Enforce the causal hierarchy: Physics → Physiology → Linguistics → Cognition → Communication
• Allow dynamic, recursive interaction between layers
• Especially connect:
• early embodiment
• later situational models

---

Table 1. The ELE Architectural Hierarchy

Stage in Continuum	Primary Scientific Domain	Operational Function / Output	Key Discrete Unit (Computational / Linguistic)	Example Empirical Measurement
Physics	Fluid Dynamics / Acoustics	Airflow generation and resonance initiation	Phonation Trigger / Acoustic Waveform	Vital Capacity (mL); Lung Pressure (cm H₂O) [12]
Physiology	Neurobiology / Motor Control	Laryngeal & articulatory muscle activation and vibration	Pneuma (Physiological / Motor Command) [14]	Cricothyroid (CT) muscle activation level [14]
Linguistics	Formal / Generative Theory	Structuring form and basic meaning	Morpheme (Minimal Meaningful Unit) [8]	ERP/MEG morphological decomposition effects [24]
Cognition	Psycholinguistics / Memory	Conceptual indexing, retrieval, and recursive structuring	Sememe (Atomic Semantic Feature) [21]	Chunking capacity (≈5–9 items); lexical retrieval time [26]
Communication	Pragmatics / Sociolinguistics	Situated contextual interpretation and intentionality	Pragmeme (Situationally Bound Speech Act) [36]	Pragmatic protocol score (topic coherence, appropriateness, etc.) [5]

---

Table 2. Cognitive and Computational Correlates of Linguistic Layers

Linguistic Level	Cognitive / Computational Correlate	Timescale / Locus (LLM / Human)	Evidence / Function
Phonology / Letter	Phonological processes; sound-to-letter mapping	Short timescale (~0.17 words); early layers [9]	Fast neurons encode basic acoustic sequence data [9]
Morphology / Word	Morphological segmentation; basic syntactic structure	Medium timescale (1–10 words); middle layers [9]	Processing minimal meaningful units and phrase structure [8]
Syntax / Phrase	Phrase-structure parsing and recursion (embedding)	Generative mechanism; RLM context management [8][10]	Enables infinite productivity via hierarchical organization [8]
Semantics / Lexical	Lexical and semantic relationships; conceptualization	Long timescale (≈360,000+ words); late layers [9]	Storage of cultural knowledge; embodied meaning indexing [3][9]
Pragmatics / Discourse	Situational adaptation; Theory of Mind inference; discourse-level control	External environment; interaction-dominant dynamics [20]	Defines meaning via context, intent, social norms, and ToM-based inference [28][39]

---

Conclusions and Architectural Recommendations

Successful implementation of the ELE requires three architectural directives:

1. Mandatory Embodiment and Sensorimotor Grounding

• Semantic units (Sememes) must be indexed through simulated physical actions or “imagined manipulation.” [15]
• Grounding is rooted in the physiological stage (mirror neuron system, embodiment) [3]
• This directly addresses critiques of ungrounded, text-only models. [6]

1. Adoption of Recursive, Multi-Scale Processing

• Use architectures like RLMs to exploit functional hierarchies (fast vs. slow neurons). [9][10]
• Handle vast contexts by recursive querying of environment and memory.
• Essential for:
  • Cognitive generalization (Metarules, chunking)
  • Pragmatic adaptation (Pragmemes, discourse integration)

1. Dynamic Interaction-Dominant Architecture

• The engine must not be a rigid pipeline.
• Empirical evidence shows situational context influences low-level processing in real time. [20]
• Therefore:
  • Physiological and early linguistic layers must remain dynamically linked to the pragmatic situational model.
  • Meaning is continuously constrained by environment and social norms (Pragmeme layer). [36]

For empirical validation, ELE must:

• Go beyond text-only corpora
• Incorporate multimodal data:
• Aerodynamic measures (VC)
• Muscle activation signals
• High-temporal-resolution neuroimaging (EEG/MEG)
• Actively correct for:
• fMRI signal loss in temporal regions
• Performance variability across groups and tasks [32]

By anchoring all formal linguistic units to measurable physical and physiological constraints, the Empirical Linguistic Engine provides a robust, human-plausible framework for next-generation computational language models.

It is:

• Physically grounded
• Biologically embodied
• Linguistically structured
• Cognitively recursive
• Communicatively situated

— a full P→P→L→C→C continuum rendered as a single coherent architecture.