;
; disambiguate-grounding.scm
;
; The word "grounding" is itself vague. Let's disambiguate it.
; What we've actually been building is CONVERGENT PROOF FROM MULTIPLE SOURCES.
;
(use-modules (srfi srfi-1))
(use-modules (ice-9 format))
(display "═══════════════════════════════════════════════════════════════════\n")
(display " DISAMBIGUATING 'GROUNDING' - What Do We Actually Mean? \n")
(display "═══════════════════════════════════════════════════════════════════\n\n")
;; ═══════════════════════════════════════════════════════════════════
;; THE VAGUENESS OF "GROUNDING"
;; ═══════════════════════════════════════════════════════════════════
(display "THE WORD 'GROUNDING' HAS MULTIPLE SENSES:\n")
(display "─────────────────────────────────────────────────────────────────\n\n")
(display " Sense 1: PHYSICAL GROUNDING (embodiment)\n")
(display " - Symbols connected to physical sensors/actuators\n")
(display " - Robot touches ball → 'ball' has physical referent\n")
(display " - Harnad's symbol grounding problem\n\n")
(display " Sense 2: EXPERIENTIAL GROUNDING (learning from world)\n")
(display " - Patterns learned from actual experience\n")
(display " - Not just told 'bank can mean X', but EXPERIENCED bank contexts\n")
(display " - Statistical patterns emerge from real usage\n\n")
(display " Sense 3: LOGICAL GROUNDING (foundation in axioms)\n")
(display " - Conclusions derived from accepted premises\n")
(display " - Mathematical proof chains back to axioms\n")
(display " - Each step justified by rules\n\n")
(display " Sense 4: EVIDENTIAL GROUNDING (convergent proof)\n")
(display " - Multiple independent sources agree\n")
(display " - Like triangulation: each source narrows possibilities\n")
(display " - THIS is what we've been building!\n\n")
;; ═══════════════════════════════════════════════════════════════════
;; WHAT WE ACTUALLY BUILT
;; ═══════════════════════════════════════════════════════════════════
(display "═══════════════════════════════════════════════════════════════════\n")
(display "WHAT WE ACTUALLY BUILT: Convergent Proof / Multi-Source Validation\n")
(display "─────────────────────────────────────────────────────────────────\n\n")
(display " Our system doesn't require a robot body.\n")
(display " It doesn't need to touch a bank to know what 'bank' means.\n\n")
(display " What it DOES do:\n")
(display " - Collect EVIDENCE from multiple independent sources\n")
(display " - Each source provides PARTIAL information\n")
(display " - Sources CONVERGE on the correct interpretation\n")
(display " - Disagreement signals uncertainty\n\n")
(display " This is TRIANGULATION / CONVERGENT PROOF:\n\n")
(display " Source 1: Co-occurrence statistics (PMI)\n")
(display " 'bank' + 'swim' → GEOGRAPHIC +4.5\n\n")
(display " Source 2: Selectional preferences (verb-object)\n")
(display " 'swim to' requires physical destination → GEOGRAPHIC +2.0\n\n")
(display " Source 3: Agent-typical actions\n")
(display " penguins swim to geographic locations → GEOGRAPHIC +3.0\n\n")
(display " Source 4: Spatial preposition semantics\n")
(display " 'to the bank' (destination) fits both\n")
(display " but 'on the bank' → GEOGRAPHIC (surface)\n\n")
(display " Source 5: Frame coherence\n")
(display " GEOGRAPHIC frame elements are activated → GEOGRAPHIC +2.0\n\n")
(display " Total: GEOGRAPHIC wins with high confidence\n")
(display " Method: PROOF BY CONVERGENCE OF INDEPENDENT EVIDENCE\n\n")
;; ═══════════════════════════════════════════════════════════════════
;; THE PROOF STRUCTURE
;; ═══════════════════════════════════════════════════════════════════
(display "═══════════════════════════════════════════════════════════════════\n")
(display "THE PROOF STRUCTURE\n")
(display "─────────────────────────────────────────────────────────────────\n\n")
(display " This is not 'sensor fusion' in the engineering sense.\n")
(display " It's closer to BAYESIAN INFERENCE / BELIEF REVISION:\n\n")
(display " ┌─────────────────────────────────────────────────────────────┐\n")
(display " │ │\n")
(display " │ Prior: P(sense) based on frequency │\n")
(display " │ │\n")
(display " │ Evidence 1: P(sense | cooccurrence) via PMI │\n")
(display " │ Evidence 2: P(sense | verb_pattern) via selectional prefs │\n")
(display " │ Evidence 3: P(sense | agent_type) via typicality │\n")
(display " │ Evidence 4: P(sense | preposition) via spatial semantics │\n")
(display " │ Evidence 5: P(sense | frame) via coherence │\n")
(display " │ │\n")
(display " │ Posterior: P(sense | all evidence) via Bayes rule │\n")
(display " │ │\n")
(display " └─────────────────────────────────────────────────────────────┘\n\n")
(display " Each evidence source is INDEPENDENT (or nearly so).\n")
(display " Convergence = multiplication of likelihoods.\n")
(display " Strong convergence = high confidence = low vagueness.\n\n")
;; ═══════════════════════════════════════════════════════════════════
;; THE ASCENDING LAYERS
;; ═══════════════════════════════════════════════════════════════════
(display "═══════════════════════════════════════════════════════════════════\n")
(display "THE ASCENDING LAYERS OF STRUCTURE\n")
(display "─────────────────────────────────────────────────────────────────\n\n")
(display " You're right - we discussed ASCENDING LAYERS earlier.\n")
(display " Each layer is built by counting patterns in the layer below.\n")
(display " Let me map our evidence sources to these layers:\n\n")
(display " Layer 0: RAW DATA\n")
(display " ─────────────────\n")
(display " - Word sequences in text\n")
(display " - (or: pixel values, audio samples)\n")
(display " - No structure yet, just raw signal\n\n")
(display " Layer 1: STATISTICAL ASSOCIATIONS (PMI)\n")
(display " ─────────────────────────────────────────\n")
(display " - Count: which words co-occur?\n")
(display " - Result: PMI table\n")
(display " - (image analog: which pixels co-vary? → textures, edges)\n\n")
(display " Layer 2: GRAMMATICAL PATTERNS (selectional preferences)\n")
(display " ─────────────────────────────────────────────────────────\n")
(display " - Count: which word categories connect how?\n")
(display " - Result: verb-object patterns, noun-adjective patterns\n")
(display " - (image analog: which shapes relate? → object parts)\n\n")
(display " Layer 3: SEMANTIC FRAMES\n")
(display " ────────────────────────\n")
(display " - Count: which patterns cluster together?\n")
(display " - Result: frames like FINANCIAL, GEOGRAPHIC, COOKING\n")
(display " - (image analog: scene categories, activity recognition)\n\n")
(display " Layer 4: AGENT MODELS (typicality)\n")
(display " ──────────────────────────────────\n")
(display " - Count: which agents do which actions?\n")
(display " - Result: penguin-typical, human-typical behaviors\n")
(display " - (image analog: person-typical poses, animal-typical habitats)\n\n")
(display " Layer 5: SPATIAL/CAUSAL STRUCTURE\n")
(display " ──────────────────────────────────\n")
(display " - Count: which prepositions go with which nouns?\n")
(display " - Result: ON=surface, IN=enclosure, BY=proximity\n")
(display " - (image analog: spatial relations between objects)\n\n")
(display " Layer 6: NARRATIVE/TEMPORAL STRUCTURE\n")
(display " ──────────────────────────────────────\n")
(display " - Count: which events follow which?\n")
(display " - Result: script expectations (restaurant script, etc.)\n")
(display " - (image analog: video action sequences)\n\n")
(display " Layer 7: SOCIAL/PRAGMATIC STRUCTURE\n")
(display " ────────────────────────────────────\n")
(display " - Count: who says what in which contexts?\n")
(display " - Result: speech acts, politeness, intent inference\n")
(display " - (image analog: social scene understanding)\n\n")
(display " Layer 8+: META-STRUCTURE\n")
(display " ────────────────────────\n")
(display " - Patterns about patterns\n")
(display " - How layers interact\n")
(display " - This is where we're working now!\n\n")
;; ═══════════════════════════════════════════════════════════════════
;; WHY SAME STRUCTURE WORKS FOR ANY MODALITY
;; ═══════════════════════════════════════════════════════════════════
(display "═══════════════════════════════════════════════════════════════════\n")
(display "WHY THE SAME STRUCTURAL PROBLEMS APPEAR IN ALL MODALITIES\n")
(display "─────────────────────────────────────────────────────────────────\n\n")
(display " The core operation is always the same:\n\n")
(display " ┌─────────────────────────────────────────────────────────────┐\n")
(display " │ │\n")
(display " │ COUNT CO-OCCURRENCES → COMPUTE STATISTICS → CLUSTER → │\n")
(display " │ → LABEL CLUSTERS → USE FOR DISCRIMINATION │\n")
(display " │ │\n")
(display " └─────────────────────────────────────────────────────────────┘\n\n")
(display " In LANGUAGE:\n")
(display " - Count: word co-occurrences\n")
(display " - Statistic: PMI\n")
(display " - Cluster: words with similar PMI profiles\n")
(display " - Labels: noun, verb, FINANCIAL, GEOGRAPHIC\n")
(display " - Discriminate: which sense fits this context?\n\n")
(display " In IMAGES:\n")
(display " - Count: pixel neighborhood co-occurrences\n")
(display " - Statistic: local contrast, gradient, texture\n")
(display " - Cluster: pixels with similar statistics\n")
(display " - Labels: edge, surface, object, background\n")
(display " - Discriminate: what object is this?\n\n")
(display " In AUDIO:\n")
(display " - Count: frequency co-occurrences over time\n")
(display " - Statistic: spectral features, formants\n")
(display " - Cluster: sounds with similar features\n")
(display " - Labels: phoneme, word, speaker, emotion\n")
(display " - Discriminate: what word was spoken?\n\n")
(display " The STRUCTURE is identical. The raw data differs.\n\n")
;; ═══════════════════════════════════════════════════════════════════
;; THE AMBIGUITY PROBLEM IS UNIVERSAL
;; ═══════════════════════════════════════════════════════════════════
(display "═══════════════════════════════════════════════════════════════════\n")
(display "THE AMBIGUITY PROBLEM IS UNIVERSAL\n")
(display "─────────────────────────────────────────────────────────────────\n\n")
(display " 'bank' has multiple senses → need to disambiguate\n\n")
(display " Visual analog:\n")
(display " Same pixel pattern can be duck or rabbit\n")
(display " Same shape can be vase or faces\n")
(display " Same colors can be dress:blue-black or dress:white-gold\n")
(display " → Need context to disambiguate!\n\n")
(display " Audio analog:\n")
(display " Same sound can be 'ba' or 'da' (McGurk effect with video)\n")
(display " Same phrase can be 'ice cream' or 'I scream'\n")
(display " Same tone can be angry or excited\n")
(display " → Need context to disambiguate!\n\n")
(display " The solution is always the same:\n")
(display " 1. Recognize the ambiguity exists\n")
(display " 2. List the possible interpretations\n")
(display " 3. Gather evidence from multiple sources\n")
(display " 4. Compute which interpretation is best supported\n")
(display " 5. Know your confidence level (vagueness)\n")
(display " 6. Seek more evidence if confidence is low\n\n")
;; ═══════════════════════════════════════════════════════════════════
;; CROSS-MODAL PROOF
;; ═══════════════════════════════════════════════════════════════════
(display "═══════════════════════════════════════════════════════════════════\n")
(display "CROSS-MODAL CONVERGENT PROOF\n")
(display "─────────────────────────────────────────────────────────────────\n\n")
(display " When modalities combine, proof becomes stronger:\n\n")
(display " Example: 'The bank is eroding'\n\n")
(display " Text alone (Layer 1-3 evidence):\n")
(display " 'eroding' has high PMI with geological terms\n")
(display " → GEOGRAPHIC sense suggested\n")
(display " Vagueness: 0.20\n\n")
(display " Add image (cross-modal evidence):\n")
(display " Image shows: river, muddy slope, water\n")
(display " Image does NOT show: building, ATM, teller\n")
(display " → GEOGRAPHIC sense CONFIRMED\n")
(display " Vagueness: 0.02\n\n")
(display " Add audio (more cross-modal evidence):\n")
(display " Sound: flowing water, birds\n")
(display " NOT: cash register, voices in queue\n")
(display " → GEOGRAPHIC sense triple-confirmed\n")
(display " Vagueness: 0.001\n\n")
(display " Each modality provides INDEPENDENT evidence.\n")
(display " Convergence multiplies confidence.\n")
(display " THIS is what 'sensor fusion' really is: CONVERGENT PROOF.\n\n")
;; ═══════════════════════════════════════════════════════════════════
;; RE-DEFINING GROUNDING
;; ═══════════════════════════════════════════════════════════════════
(display "═══════════════════════════════════════════════════════════════════\n")
(display "RE-DEFINING GROUNDING: Convergent Evidence Across Layers\n")
(display "─────────────────────────────────────────────────────────────────\n\n")
(display " Let's replace the vague term 'grounding' with precise terms:\n\n")
(display " OLD: 'The system grounds the word in reality'\n")
(display " NEW: 'The system has CONVERGENT EVIDENCE from multiple\n")
(display " LAYERS and SOURCES that this interpretation is correct'\n\n")
(display " OLD: 'The meaning is grounded'\n")
(display " NEW: 'The vagueness is low because evidence converges'\n\n")
(display " OLD: 'Embodied grounding'\n")
(display " NEW: 'Evidence from sensorimotor layers in addition to\n")
(display " symbolic layers, all converging'\n\n")
(display " The key insight:\n\n")
(display " ┌─────────────────────────────────────────────────────────────┐\n")
(display " │ │\n")
(display " │ 'GROUNDING' = Low vagueness achieved through │\n")
(display " │ convergent evidence from multiple layers │\n")
(display " │ │\n")
(display " │ More layers + more agreement = more 'grounded' │\n")
(display " │ │\n")
(display " │ Physical embodiment is ONE source of layers, │\n")
(display " │ but not the only one. Text-only systems can be │\n")
(display " │ 'grounded' through statistical/structural convergence. │\n")
(display " │ │\n")
(display " └─────────────────────────────────────────────────────────────┘\n\n")
;; ═══════════════════════════════════════════════════════════════════
;; BUILDING HIGHER LAYERS
;; ═══════════════════════════════════════════════════════════════════
(display "═══════════════════════════════════════════════════════════════════\n")
(display "BUILDING HIGHER LAYERS: Counting Structure In Structure\n")
(display "─────────────────────────────────────────────────────────────────\n\n")
(display " The system must not only HAVE layers, but DISCOVER them.\n")
(display " How? By counting patterns in existing structure.\n\n")
(display " Level N+1 = patterns found by counting Level N patterns\n\n")
(display " Example chain:\n")
(display " L0: Words ('the', 'bank', 'eroded')\n")
(display " L1: Co-occurrences ('bank-eroded' has PMI=3.2)\n")
(display " L2: Categories (words with similar PMI profiles cluster)\n")
(display " L3: Frames (category combinations that co-occur)\n")
(display " L4: Scripts (frame sequences over time)\n")
(display " L5: Meta-patterns (which scripts share structure?)\n\n")
(display " Each level is discovered by the SAME operation:\n")
(display " Count → Statistics → Cluster → Label\n\n")
(display " The system builds its own hierarchy by recursively applying\n")
(display " the same structural analysis to its own outputs.\n\n")
;; ═══════════════════════════════════════════════════════════════════
;; SUMMARY
;; ═══════════════════════════════════════════════════════════════════
(display "═══════════════════════════════════════════════════════════════════\n")
(display "SUMMARY: What We're Really Building\n")
(display "─────────────────────────────────────────────────────────────────\n\n")
(display " 1. Not 'grounding' in the fuzzy sense\n")
(display " → CONVERGENT PROOF from multiple evidence layers\n\n")
(display " 2. Not 'sensor fusion' in the engineering sense\n")
(display " → BAYESIAN BELIEF UPDATE with independent sources\n\n")
(display " 3. Not just 'statistics'\n")
(display " → RECURSIVE STRUCTURE DISCOVERY (counting counts)\n\n")
(display " 4. The same method works for any modality\n")
(display " → Count → Cluster → Label → Discriminate\n\n")
(display " 5. Higher layers emerge from counting lower layers\n")
(display " → Autonomous structure building\n\n")
(display " 6. 'Knowing' = having enough layers with enough convergence\n")
(display " → Vagueness approaches zero\n\n")
(display " The path forward:\n")
(display " - Keep building layers\n")
(display " - Add more evidence sources\n")
(display " - Maintain convergence checking\n")
(display " - Let the system discover new layers autonomously\n\n")