From 'Feces' to 'Breath': Two Hidden Fronts of Structure Recovery
> One-line summary: Two papers, one soul — both ask: how do we recover lost historical structure from messy modern data? One recovers how Africans said 'feces' thousands of years ago; the other recovers subtle distortions in your breathing when you're fatigued.
Opening: Two Seemingly Unrelated Worlds
The first paper: *Neural Recovery of Historical Lexical Structure in Bantu Languages from Modern Data* — using neural networks to recover the historical lexical structure of Bantu languages.
The second paper: *Time-Localized Parametric Decomposition of Respiratory Airflow for Sub-Breath Analysis* — decomposing a single breath into several parametric wave components.
On the surface: one is linguistics, one is physiology. In reality: both do the same thing — reconstruct structured truth from chaotic, irreversible modern observations where information has been lost.
Act I: Historical Code Hidden in 'Feces'
Background
The Bantu language family is one of Africa's largest, spoken by ~300 million people. Most Bantu languages have no written records. Linguists want to know what Proto-Bantu looked like before these languages split, roughly 3000–5000 years ago — before writing existed.
The Traditional Method
The Comparative Method: 1. Find cognates (e.g., English *father*, German *Vater*, Latin *pater*) 2. List sound change laws (e.g., p → f is common in Germanic) 3. Reconstruct ancestral forms
Problem: relies on linguists' manual expertise, extremely time-consuming, and prone to missing patterns.
The Paper's Solution: Let Neural Networks 'Learn' Sound Changes
Core idea: turn ancestral reconstruction into a sequence-to-sequence translation problem.
| Traditional method | This paper | |---------|--------------| | Manual cognate comparison | Deep sequence models learn automatically | | Manually summarized sound laws | Model extracts implicit patterns from data | | One word at a time | Large-scale parallel processing of whole lexicons |
Details:
- Transformer architecture (same family as ChatGPT)
- Input: modern word forms (e.g., Swahili *-taka*)
- Output: reconstructed ancestral form (e.g., *-tákà)
- Training data: East Bantu cognate sets with lexicalized verb roots
- Tidal Volume
- Inspiratory Time
- Peak Flow
- Symmetry Index
- Gaussian model: most stable parameters (std < 0.04), but largest reconstruction error
- Beta model: smallest reconstruction error, but least stable shape parameters (α, β) (std 1.0+)
- Half-sine model: best compromise — error near Beta, stability near Gaussian
- Paper 1: Neural Recovery of Historical Lexical Structure in Bantu Languages from Modern Data
- arXiv: 2604.22730v1 [cs.CL]
- Authors: N. Abdou, L. Lee, A. Lim, M. S. Seck
- Paper 2: Time-Localized Parametric Decomposition of Respiratory Airflow for Sub-Breath Analysis
- arXiv: 2604.22695v1 [eess.SP]
- Authors: Victoria Ribeiro Rodrigues, Paul W. Davenport, Nicholas J. Napoli
- Funding: Office of Naval Research (N00014-22-1-2653)
The Most Interesting Detail: Why 'Feces'?
The paper notes:
> "...the Bantu lexicon for taboo bodily functions, such as those related to feces, provides a particularly rich test case due to..."
Because taboo words: 1. Drift semantically fast: people avoid saying them directly, constantly inventing euphemisms 2. Are culturally specific: societies handle taboos differently 3. Have complex cognate relations: one root may be inherited, replaced, or borrowed differently across daughter languages
In other words, 'feces' is a stress test for language evolution — if the model can reconstruct ancestors from these chaotic taboo words, ordinary vocabulary is easy. It's like learning to drive not on an empty straight road, but in a rush-hour roundabout.
Technical Innovation: Pattern Attention
Standard Transformer attention is global. But sound changes are usually local (e.g., a vowel rises after certain consonants). The paper proposes pattern attention: the model learns which local patterns trigger which sound change rules — similar to how human linguists notice rules like 'vowels raise after bilabials.'
Act II: The 'Proto-Language' Inside Breath
Why Respiratory Signals Are Hard
> A breath is not one event but a composite action of multiple muscle groups.
The diaphragm, external intercostals, accessory muscles... each activates at different times, contributing differently shaped flow waveforms that superimpose into the observed signal.
The Traditional Approach: Blunt 'One-Size-Fits-All'
Global metrics only:
Problem: like measuring only a painting's width and height while ignoring what the painting depicts.
The Solution: Decomposing Breath into 'LEGO Blocks'
Core idea: model each breath as a superposition of a few time-localized components, each defined by: 1. Amplitude — how much flow it contributes 2. Onset Time — when it starts 3. Duration — how long it lasts 4. Shape parameters — the waveform's specific form
Three candidate basis functions:
| Basis | Shape | Pros | Cons | |--------|------|------|------| | Gaussian | Bell curve | Mathematically stable, few parameters | Can't describe sharp changes | | Half-Sine | Half sine wave | Clear start/end, physiologically intuitive | Too symmetric | | Beta | Tunable skew | Most flexible, handles asymmetry | Shape parameters hard to interpret |
Optimization: constrained nonlinear fitting of each waveform as a sum of components.
The Most Interesting Finding: Fatigue Changes Breath's 'Grammar'
Classification experiment: distinguishing 'fatigued' vs. 'non-fatigued' states.
| Feature set | Matthews Correlation Coefficient (MCC) | |--------|-------------------------| | Traditional global features (2) | 0.419 | | Traditional global features (4) | 0.414 | | Component-level features (2) | 0.482 (+15%) | | Component-level features (4) | 0.541 (+30.7%) |
Key insight: fatigue doesn't just make breathing shallower or slower — it changes the temporal coordination of internal respiratory components. As cognitive fatigue develops, breathing shifts from 'automatic mode' to 'cognitively controlled mode,' desynchronizing muscle group activation — like an orchestra where players who normally play with eyes closed must now stare at the conductor, losing the rhythm.
Noise Robustness
Test: 30dB Gaussian white noise added, fitting repeated 50 times.
Results:
Analogy: Gaussian is a prime lens (stable, reliable, inflexible); Beta is a big zoom lens (versatile, variable quality); half-sine is a quality standard zoom (capable, stable, good image quality).
Act III: The Common Structure of Two Fronts
Core Problem: Information Is Lost — How to Recover It?
| | Paper 1: Ancestral reconstruction | Paper 2: Breath decomposition | |---|--------------|----------------| | Lost information | Proto-Bantu phonological structure | Independent contributions of respiratory muscles | | Observed | Modern word form variants | Superimposed airflow waveform | | Irreversibility | Sound change is one-way, random | Mixed muscle activity can't be directly separated | | Reconstruction target | Ancestral root forms | Component parameters (amplitude, timing, shape) | | Constraints | Known sound laws, morphological rules | Physiological plausibility (amplitude > 0, time > 0) |
Shared Methodological Traits
1. Parameterized implicit models: Paper 1's Transformer encodes sound laws implicitly in attention weights; Paper 2's basis function parameters encode physiological components 2. Constrained optimization: linguistic constraints (phoneme inventory, syllable structure) vs. physiological constraints (non-negativity, temporal ordering) 3. Non-uniqueness: one modern form may map to multiple historical paths; different component combinations may produce similar waveforms 4. Physical plausibility checks: reconstructed forms must match linguistic typology; component parameters must match respiratory physiology
Both Reveal the Importance of 'Internal Structure'
Paper 1: global comparison misses local sound change patterns — only learning which local contexts trigger which changes enables accurate reconstruction.
Paper 2: global breathing metrics miss internal temporal structure — only decomposing component timing coordination detects fatigue's subtle effects.
Shared lesson: Aggregated metrics lose information. Understanding a system requires entering its internal structure.
Act IV: What Would Feynman Say?
> "You're both solving the same problem — finding the original thread in a tangled mess. One mess is linguistic, one is physiological. Different methods, same mindset. > > The first paper reminds me of a nautilus shell: it grows ring by ring, each ring recording its childhood shape. Modern languages are the outermost ring; the proto-language is the innermost. Linguists used to use magnifying glasses, comparing distances between rings. Now they use X-rays — neural networks see through several layers at once, processing thousands of shells. > > But their best idea wasn't the Transformer — it was choosing 'feces.' Why? Because it's hard. Chaotic. People avoid saying it directly, so variants are maximal and irregular. If you can find patterns in the messiest cases, you truly understand the pattern. > > The second paper reminds me of spectroscopy. How do we know what's in the sun? Not by looking directly — too bright. We look at which frequencies are absorbed. Each element has a spectral 'fingerprint.' > > Breathing is the same. The waveform is a sum of many muscles working at once, like many elements' light mixed together. This paper assumes each muscle's activity is a 'basis function' — half-sine, Gaussian, or Beta — then asks: how many such fingerprints, at what intensity, at what times, best explain the observed waveform? > > Their smartest finding: fatigue changes not how deep you breathe, but the timing relations between 'fingerprints.' Like an orchestra — fatigue doesn't weaken each player; they drift apart in rhythm. > > Both papers do reverse engineering. One reverses language history; the other reverses physiological mechanism. The tools differ, but the core question is shared: observations are mixed, information is lost, the process is irreversible — what do you do? > > My answer: find constraints. Physical, physiological, linguistic. More constraints means a smaller solution space and more reliable reconstruction. > > Paper 1's constraint: sound change isn't arbitrary — it follows human speech perception and articulatory physiology. Paper 2's constraint: muscle activation order isn't arbitrary — it follows neural control hierarchy. > > Without constraints, you're guessing. With constraints, you're solving a puzzle."
Conclusion: The Science of Structure Recovery
These papers seem unrelated, but both touch one of science's deepest questions:
> When direct observation has irreversibly lost structural information, how do we reconstruct that lost world?
The linguist's answer: exploit the regularity of sound change.
The physiologist's answer: exploit the constraints of muscle activation.
Two answers, one belief: beneath chaotic surfaces, there is always structure waiting to be discovered.