PATIENT VERBAL TELEMETRY:

Standardized Expressive Elicitation as Longitudinal Clinical Data

John Swygert
July 15, 2026
DOI: To be assigned

Abstract

Clinical medicine routinely measures pulse, blood pressure, temperature, oxygen saturation, laboratory chemistry, electrical activity, movement, sleep, and anatomical structure. Yet one of the richest signals continuously produced by a patient—the organization of their language—is often reduced to a brief provider summary or scattered quotations in a medical record.

This paper proposes Patient Verbal Telemetry, a standardized method for collecting, preserving, analyzing, and longitudinally comparing patient-generated spoken and written language. “Verbal” is used here in its broad sense of expression through words and includes speech, dictated narrative, prose, poetry, lyrics, structured responses, and other language-based samples.

The framework does not propose that words alone can establish a diagnosis, determine dangerousness, or reveal a person’s intentions with certainty. Patient expression may be affected by culture, dialect, education, personality, performance, fiction, concealment, medication, fatigue, pain, neurological condition, provider bias, and the conditions under which the sample was produced. Verbal telemetry must therefore remain one signal among many and must be interpreted through personal baseline, clinical context, corroborating evidence, and human review.

The central proposal is that language samples should be elicited through repeatable tasks, preserved in their original form, accompanied by contextual metadata, and compared over time. Features may include semantic content, narrative coherence, acoustic properties, self-reference, agency, temporal orientation, cognitive flexibility, perspective-taking, emotional range, route diversity, and changes from the patient’s own established baseline.

Expressive writing may also have a dual role. It can provide a patient with an outlet for organizing experience while simultaneously producing a clinically useful record of how that experience is being represented. Research on expressive writing suggests benefits in some populations and circumstances, but findings are mixed and do not support treating it as a universal therapy.

The fundamental proposition is:

The patient has always been transmitting verbal telemetry. Clinical science has lacked a standardized, ethical, longitudinal method for receiving, preserving, comparing, and integrating it.

1. Introduction

A patient enters a medical office and generates an enormous amount of information.

They describe pain.

They explain what happened.

They hesitate.

They repeat particular themes.

They avoid others.

They move between past, present, and future.

They construct causes.

They assign responsibility.

They imagine—or fail to imagine—alternatives.

They speak with speed, slowing, tension, flattening, humor, uncertainty, anger, distance, coherence, fragmentation, or exhaustion.

Most of this information is transient.

The clinician may preserve a few observations:

patient appears anxious;
speech is pressured;
mood is depressed;
thought process is circumstantial;
patient denies suicidal intent;
patient reports increased pain.

These descriptions may be clinically valuable, but they are already processed through the observer.

The patient’s original expression is usually lost.

Modern medicine increasingly recognizes that speech and language may contain measurable information relevant to cognition, neurological disease, depression, psychosis, physical aging, and treatment response. Recent studies have investigated longitudinal speech changes in psychosis, cognitive decline, depression, Parkinson’s disease, amyotrophic lateral sclerosis, and aging. The results are promising but remain heterogeneous, context-dependent, and insufficient for stand-alone diagnosis.

The missing step is not merely a more powerful classifier.

It is a complete clinical collection architecture.

That architecture must answer:

What language should be collected?
Under what conditions?
How frequently?
What contextual information must accompany it?
Which features are useful?
How should change be measured?
How can provider and algorithmic bias be restrained?
Who may access the material?
How should the patient participate in interpretation?
What clinical decisions, if any, may legitimately follow?

Patient Verbal Telemetry is proposed as a framework for answering those questions.

2. Definition

Patient Verbal Telemetry is the ethical, repeatable, context-documented collection of patient-generated language for longitudinal clinical comparison and integration with other health information.

The framework includes both spoken and written expression.

Possible sources include:

ordinary clinical conversation;
structured oral prompts;
recorded voice diaries;
dictated narratives;
handwritten or typed responses;
poetry;
lyrics;
descriptions of images or events;
perspective-taking tasks;
future-oriented writing;
and patient-selected creative work.

The word telemetry does not imply constant surveillance.

It refers to repeated signals produced by a changing system and collected so that variation can be studied over time.

The patient is the system being observed, but not reduced to the signal.

The language sample is evidence, not identity.

3. Why Language Is Clinically Valuable

Language sits at the intersection of multiple systems.

Producing a sentence may involve:

attention;
memory;
motor planning;
respiration;
articulation;
vocabulary access;
syntax;
emotional state;
social awareness;
narrative organization;
inhibition;
interpretation;
and prediction.

This makes language unusually information-rich.

A change in speech may reflect:

neurological impairment;
medication effects;
fatigue;
respiratory limitation;
pain;
emotional distress;
cognitive decline;
intoxication;
sleep loss;
social withdrawal;
or changes in thought organization.

The same observed feature may have several possible causes.

Slowed speech, for example, may arise from depression, motor disease, sedation, exhaustion, pain, deliberate caution, or normal variation.

Therefore, verbal telemetry should not be interpreted through a one-feature-to-one-diagnosis model.

Its value lies in relationships among features and in changes over time.

Research following people with psychotic disorders has found longitudinal relationships between clinical symptom domains and combinations of automated speech and language features across repeated assessments. The researchers described the long-term objective as developing rapidly obtainable “vital signs” of psychosis severity, while also emphasizing the limited number of longitudinal studies currently available.

That analogy is useful, but verbal telemetry differs from blood pressure.

Blood pressure has standardized units, mature reference ranges, and comparatively direct physical interpretation.

Language is culturally embedded, strategically used, genre-dependent, and highly contextual.

It may become a clinical vital signal, but never a context-free one.

4. Expression Is Not Confession

A fundamental safeguard is necessary:

Patient-generated language must not automatically be treated as literal confession, factual reporting, or transparent access to intention.

A patient may:

write fiction;
inhabit a character;
exaggerate for artistic effect;
minimize distress;
conceal an intention;
imitate a style;
use dark humor;
experiment with language;
comply performatively with a clinician;
or write what they believe the clinician wants to hear.

A violent poem does not establish violent intent.

A hopeful poem does not establish safety.

Frequent first-person language does not establish narcissism.

Fragmented syntax does not by itself establish thought disorder.

A lyric about death does not necessarily represent suicidal planning.

Expression must be interpreted through:

the task;
the genre;
the patient’s explanation;
the patient’s history;
accompanying behavior;
direct clinical assessment;
and change over time.

The proposed method studies expression without pretending that expression is perfectly transparent.

5. Expressive Writing as Outlet and Measurement

Writing can perform two functions simultaneously.

First, it may help the patient externalize and organize experience.

Second, it produces an artifact that can be preserved and later compared.

The act of writing may change the internal state being measured. That does not invalidate the sample.

It means the sample is both:

an observation;
and a possible intervention.

This is common in medicine.

A physical examination may change pain.

A diagnostic interview may change understanding.

A mobility test may produce fatigue.

The observer and the procedure participate in the event.

Research on expressive writing has reported small or condition-specific benefits, including improvements in some measures of stress, trauma-related symptoms, depression, working memory, or health. Other trials have found little broad benefit or outcomes that depend upon individual expressiveness and context. The evidence therefore supports studying writing as a potentially useful supplement, not declaring it universally therapeutic.

A patient who finds writing distressing should not be forced to continue.

A patient who benefits from visual art, movement, conversation, music, or silence should not be told that writing is the only legitimate route.

Patient Verbal Telemetry is a route, not the only route.

6. Standardized Expressive Elicitation

Casual writing can be clinically informative, but scientific comparison requires some standardization.

A Standardized Expressive Elicitation, or SEE, is a language-producing task administered under documented and reasonably repeatable conditions.

A task might ask:

Write for ten minutes about the event that has occupied your thoughts most strongly this week.

Another might ask:

Describe what happened, how you interpreted it, and what you believe may happen next.

A third might ask:

Describe the same event from the perspective of another person who was present.

A fourth:

List three different explanations for why this event occurred.

A fifth:

Write about the experience without using the words “I,” “me,” or “my.”

A sixth:

Describe one action you could take, one action someone else could take, and one possibility neither of you currently controls.

These tasks do not diagnose conditions.

They elicit different forms of organization.

They may provide information about:

emotional access;
self-reference;
causal reasoning;
perspective-taking;
cognitive flexibility;
anticipated future;
perceived agency;
alternative-route generation;
narrative coherence;
and relationship between internal and external focus.

7. The Elicitation Record

Every sample should be stored with its production context.

At minimum, the record should include:

exact prompt;
date and time;
duration;
spoken or written format;
location;
language and dialect;
whether assistance was provided;
whether the patient edited the response;
current pain level;
sleep quantity or major sleep disruption;
relevant medication timing;
recent substance use when clinically appropriate;
acute stressors;
sensory or motor limitations;
and the patient’s description of what the response means.

The patient’s own interpretation is essential.

The clinician may see anger.

The patient may say the passage was satire.

The algorithm may detect hopeless language.

The patient may explain that the poem was written from the perspective of a deceased relative.

Those interpretations may still require exploration, but none should silently replace the original source.

8. Preserve the Original Signal

The unaltered sample should be retained whenever the patient has consented to retention.

For writing, this may mean:

an image of the handwritten page;
the original digital file;
edit history when ethically and technically appropriate;
and a transcription for analysis.

For speech, it may mean:

original audio;
a time-aligned transcript;
and derived acoustic features.

The record should distinguish among:

Original patient expression
Patient explanation
Clinician observation
Automated analysis
Clinical interpretation
Later outcome

These layers should never be silently merged.

A sentence such as “patient demonstrates persecutory thinking” is an interpretation.

The original words that led to that interpretation should remain inspectable by authorized clinicians whenever practicable.

This reduces the chance that one provider’s language becomes an unchallengeable boundary around the patient’s future care.

9. Layers of Verbal Telemetry

Verbal telemetry can be divided into several analytic layers.

9.1 Lexical Layer

This includes word choice and vocabulary characteristics such as:

first-person references;
social references;
threat-related language;
certainty terms;
emotional vocabulary;
causal language;
future-oriented language;
absolutist words;
negation;
and lexical diversity.

No lexical feature is inherently pathological.

Its meaning depends on task, context, baseline, and interaction with other features.

9.2 Syntactic Layer

This includes:

sentence length;
grammatical complexity;
subordinate clauses;
incomplete constructions;
repetition;
and changes in structural organization.

Recent studies have associated patterns such as lexical-syntactic simplification with aspects of aging or physical function, but these remain emerging associations rather than universally established diagnostic rules.

9.3 Semantic Layer

This concerns meaning relationships.

Possible features include:

topic recurrence;
semantic distance;
coherence between sentences;
unexpected shifts;
causal continuity;
contradiction;
fixation;
and the number of distinct interpretations generated.

Research has investigated semantic coherence and other language features in psychosis, but translating these measurements into responsible clinical tools remains difficult.

9.4 Narrative Layer

This includes:

beginning, development, and resolution;
continuity of identity;
assignment of agency;
perceived cause;
anticipated future;
and whether the narrative permits change.

A narrative may repeatedly end in:

withdrawal;
retaliation;
helplessness;
reconciliation;
service;
escape;
or uncertainty.

The repeated endpoint may be clinically informative, but it is not automatically causal.

9.5 Acoustic Layer

For spoken samples, features may include:

speech rate;
pause duration;
pitch range;
loudness;
articulation;
voice quality;
rhythm;
breath patterns;
and spectral characteristics.

These features may contain information about motor function, mood, respiratory condition, neurological disease, fatigue, and other variables. Current research is exploring their use in depression, cognitive monitoring, ALS, Parkinson’s disease, and aging.

9.6 Interaction Layer

In conversation, additional features arise:

response latency;
interruption;
repair;
turn length;
responsiveness;
topic maintenance;
reciprocity;
and adaptation to another speaker.

This layer is absent from solitary writing and should be analyzed separately.

9.7 Creative Layer

Poetry, song lyrics, metaphor, and fiction contain information not captured by literal content alone.

Possible features include:

image selection;
emotional contrast;
perspective;
recurring symbols;
tonal shifts;
metaphorical distance;
and the relationship between form and content.

Creative work must be handled especially carefully because artistic voice may differ substantially from autobiographical voice.

10. Personal Baseline Before Population Label

The most valuable comparison may not be:

Does this person sound like the average person with diagnosis X?

It may be:

How does this person differ from themselves six months ago?

Each person has a characteristic range of:

vocabulary;
humor;
tempo;
sentence structure;
emotional expression;
abstraction;
self-reference;
social reference;
and narrative style.

A personal baseline can be built from repeated samples collected while the person is comparatively stable.

Let the feature representation of a sample at time be:


\mathbf{x}_t =
\left[
L_t,
S_t,
M_t,
N_t,
A_t,
C_t
\right]

where:

represents lexical features;
represents syntactic features;
represents semantic features;
represents narrative features;
represents acoustic features;
represents contextual variables.

The clinically useful quantity may be less the absolute value of than its deviation from the patient’s established range:


\Delta_t = \mathbf{x}_t-\boldsymbol{\mu}_{\text{patient}}

where represents the personal baseline.

This is a conceptual model, not a validated clinical formula.

A large deviation should produce a question, not a verdict:

Something has changed. What explains it?

Possible explanations may include:

worsening symptoms;
improvement;
medication change;
pain;
sleep deprivation;
grief;
a new relationship;
intoxication;
neurological change;
deliberate artistic experimentation;
or simple day-to-day variation.

11. Route-Space Analysis

Within TSTOEAO, patient expression can be examined as evidence of the routes through which experience is being interpreted.

A simplified sequence is:


\text{Experience}
\rightarrow
\text{Emotional Gradient}
\rightarrow
\text{Interpretive Boundary}
\rightarrow
\text{Available Thought Routes}
\rightarrow
\text{Verbal Expression}
\rightarrow
\text{Behavioral Resolution}

Consider the statement:

Nothing I do ever works.

The words may indicate several possible structures:

Gradient: disappointment, exhaustion, shame, fear, or pain.

Interpretive boundary: prior failures are being treated as evidence governing every future attempt.

Suppressed routes: partial success, assistance, experimentation, delay, adaptation, or changed conditions.

Dominant route: withdrawal or non-action.

Expected resolution: future attempts are anticipated as failures before they occur.

The clinician should not assume this entire structure from one sentence.

The sentence suggests questions:

What has not worked?
Has anything worked partially?
What changed between attempts?
What would count as success?
Are there actions the patient has not considered?
Is the person describing objective constraint, emotional exhaustion, or both?

Route-space analysis converts language into a map for inquiry rather than a label.

12. The Route Portfolio of Thought

A psychologically flexible narrative often contains several possible continuations.

A constricted narrative may repeatedly produce only one.

Potential verbal indicators of route constriction include:

only one causal explanation;
only one anticipated outcome;
repeated inevitability;
inability to imagine another person’s perspective;
absence of future possibilities;
absolute self-blame;
absolute external blame;
repeated return to a single grievance;
or inability to propose any tolerable action.

These features are not diagnoses.

Some situations genuinely contain very few available routes.

A person in severe poverty, confinement, abuse, terminal illness, or legal jeopardy may accurately describe major external restriction.

Clinical analysis must not convert real constraint into presumed cognitive distortion.

The proper question is:

Which routes are objectively unavailable, which are emotionally inaccessible, and which might be reopened through treatment, resources, protection, or changed conditions?

Treatment may then involve route portfolio repair:

establishing physical safety;
reducing pain;
adding social support;
challenging a rigid interpretation;
creating one low-risk action;
practicing perspective shift;
identifying a future state;
or relocating an intolerable gradient through safe expression.

13. A Proposed Route Flexibility Profile

A future research instrument could examine several dimensions.

13.1 Interpretive Diversity

How many plausible explanations can the patient generate?

13.2 Action Diversity

How many possible responses can the patient imagine?

13.3 Perspective Mobility

Can the event be represented from another viewpoint without losing contact with the patient’s own experience?

13.4 Temporal Openness

Does the future appear fixed, absent, threatening, hopeful, or uncertain?

13.5 Agency Distribution

Is all agency assigned to the self, to other people, to institutions, to chance, or to no one?

13.6 Boundary Rigidity

Does new evidence modify the narrative, or is every new event forced into the same explanation?

13.7 Emotional Range

Can several emotions coexist, or is the narrative dominated by one undifferentiated state?

13.8 External Reality Contact

Are other people represented as independent persons, or only as threats, judges, rescuers, instruments, or mirrors?

13.9 Resolution Pattern

Where do narratives repeatedly end?

action;
service;
repair;
surrender;
retaliation;
withdrawal;
confusion;
or openness.

No single dimension should be used as a diagnostic score without substantial validation.

The profile is a research proposal for organizing inquiry.

14. Structured Creative Tasks

Creative writing may permit expression that ordinary clinical questioning does not.

A patient may resist:

Tell me how angry you are.

But respond richly to:

Write a short scene in which anger enters a room as a person.

A patient may be unable to answer:

Do you feel hopeful?

But may reveal temporal orientation through:

Write a letter from yourself one year in the future.

Possible standardized creative tasks include:

The Unedited Emotional Narrative

Write continuously about the most emotionally significant event of the week.

The Alternate Explanation Task

Write three different explanations for the same event.

The Other-Person Task

Write the event from the viewpoint of another person.

The Future-Self Task

Write from a future point in which the current problem has changed.

The Route Inventory

List every response available, including responses you would reject.

The Boundary Task

Describe what feels impossible, forbidden, unsafe, or unreachable.

The External-Focus Task

Write about the event without describing yourself.

The Metaphor Task

Represent the present emotional state as a place, weather system, machine, room, animal, or landscape.

The Service Task

Describe one way the experience might help another person.

These tasks could reveal different capacities without requiring that patients use clinical terminology.

15. Music, Lyrics, and Multilayer Expression

Lyrics add another level of complexity.

A song contains multiple interacting channels:

literal words;
vocal delivery;
melody;
harmony;
rhythm;
tempo;
timbre;
instrumentation;
dynamics;
production;
silence;
and temporal development.

The emotional meaning of a song cannot always be inferred from lyrics alone.

Somber words may be paired with energetic music.

Hopeful language may appear over unstable harmony.

A calm vocal delivery may carry severe content.

A title may transform the interpretation of an otherwise ambiguous lyric.

The original listening experience should therefore be preserved before separation.

A possible analytic sequence is:


\text{Whole Song}
\rightarrow
\text{Vocal Isolation}
\rightarrow
\text{Lyric Verification}
\rightarrow
\text{Instrumental Separation}
\rightarrow
\text{Layer Analysis}
\rightarrow
\text{Reintegration}

This has direct relevance to verbal telemetry.

Words are often embedded in acoustic context.

A listener may mishear them because vowels are stretched, consonants are masked, accents differ, reverb smears boundaries, and instruments occupy overlapping frequencies.

The complete signal must be experienced as a whole, but uncertain language should be verified before semantic interpretation is treated as reliable.

16. Clinical Integration

Verbal telemetry should never operate in isolation.

A clinical interpretation may integrate:


\text{Verbal Telemetry}
+
\text{Medical History}
+
\text{Physical Examination}
+
\text{Laboratory Data}
+
\text{Medication Data}
+
\text{Behavioral Observation}
+
\text{Patient Report}
+
\text{Collateral Information}

The weighting will vary by use case.

For neurological monitoring, acoustic and motor-speech features may receive greater emphasis.

For psychotherapy, narrative organization and personal meaning may matter more.

For pain medicine, verbal samples may help track function, coping, sleep, perceived control, and treatment response.

For rehabilitation, speech and writing may reveal changes in motor control, cognition, confidence, and participation.

For psychiatry, language may help identify change in affect, organization, flexibility, social connection, and symptom burden.

No model should silently convert one domain’s findings into another domain’s diagnosis.

17. Alerts, Not Verdicts

A responsible system should produce statements such as:

The patient’s language has become more repetitive, future-closed, and self-blaming relative to their previous six samples.

Or:

Pause duration and speech rate have changed substantially from this patient’s baseline. Pain, fatigue, medication effects, mood, respiratory function, and neurological causes should be considered.

It should not automatically declare:

The patient is depressed.

The patient is psychotic.

The patient is manipulative.

The patient is dangerous.

The patient will become violent.

A telemetry system detects signal changes.

Clinical professionals investigate causes.

Current speech-AI research repeatedly emphasizes challenges involving small datasets, limited generalizability, explainability, task differences, and translation from research settings into clinical care.

18. Violence, Stigma, and Misuse

Patient Verbal Telemetry must not become a system for labeling suffering people as future criminals.

Most distress does not become violence.

Most unusual language does not represent dangerousness.

Most people with psychiatric illness are not violent.

Risk assessment requires direct inquiry, history, behavior, environmental conditions, substance use, access, threats, protective factors, and professional judgment.

Creative writing may contain:

rage;
fantasy;
revenge;
death;
destruction;
or fear.

Such content may warrant respectful exploration.

It should not automatically trigger punishment, police involvement, loss of employment, insurance consequences, or permanent stigmatizing labels.

The purpose is earlier understanding and more appropriate support—not preemptive social exclusion.

When a patient communicates a specific and credible threat or immediate danger, clinicians must follow applicable professional and legal duties. But no generalized automated language score should substitute for that assessment.

19. Manipulation and Unreliable Reporting

The patient is not the only possible source of distortion.

Providers may be:

biased;
hurried;
culturally uninformed;
punitive;
overconfident;
financially motivated;
institutionally pressured;
or intentionally manipulative.

Algorithms may reproduce bias from their training data.

Institutions may misuse ambiguous findings.

Therefore, the design must assume that every interpretive layer can fail.

Protection requires:

preservation of original samples;
clear separation of observation and interpretation;
patient access where legally appropriate;
documentation of model version;
uncertainty reporting;
independent audit;
appeal and correction;
and prohibition against unauditable black-box conclusions in high-stakes decisions.

NIST warns that AI systems may scale bias, remove necessary context from complex human phenomena, and invite excessive deference because users perceive automated outputs as objective.

Patient verbal telemetry should be designed to expose interpretation—not hide it.

20. Consent

Patients should understand that their speech or writing will be preserved and analyzed.

They do not necessarily need to be told every individual feature that may later be measured before each task, because excessive detail could alter the response.

But they should be told the fundamental purpose.

A consent explanation might state:

I would like you to complete a brief spoken or written exercise. With your permission, we will preserve it in a protected part of your health record and compare it with future samples as one source of information about how you are doing. It will not be used by itself to diagnose you or determine that you are dangerous. You may ask questions, decline, or discuss what the sample means to you.

Separate permission should be obtained for:

clinical care;
research;
model development;
publication;
education;
and commercial use.

Consent for treatment should not be treated as unlimited consent to train artificial intelligence systems.

21. Privacy and the Sacred Record

The intimate nature of patient expression requires stronger protection than ordinary consumer data.

A poem about grief, a voice diary recorded during crisis, or a narrative about trauma may expose more than a laboratory result.

In the United States, HIPAA generally treats most mental-health information under the ordinary Privacy Rule, while separately maintained psychotherapy notes receive additional protections. Not every expressive sample placed in a medical record would automatically qualify as a psychotherapy note.

Therefore, the framework should not rely solely on existing minimum legal categories.

It should propose an Enhanced Expressive Record with:

restricted access;
access logging;
encryption;
prohibition on commercial advertising use;
strict secondary-use consent;
limited export into general chart summaries;
patient notification of research use;
and retention rules appropriate to clinical purpose.

The record is sacred not because it can never be used, but because every use must serve a legitimate and disclosed purpose.

22. Data Standardization

A clinically useful system requires common definitions and collection procedures.

The FDA’s guidance on digital health technologies emphasizes fit-for-purpose technologies, reliable data acquisition, participant usability, record integrity, and appropriate handling of remotely collected data in clinical investigations.

A verbal telemetry standard should specify:

microphone characteristics;
recording format;
noise conditions;
transcription method;
prompt wording;
task duration;
language;
translation method;
missing-data handling;
metadata;
model version;
feature definitions;
and quality-control procedures.

Written tasks should document:

handwriting versus typing;
autocorrect;
spell-check;
dictation;
editing permissions;
time allowed;
and outside assistance.

Without standardization, apparent clinical differences may be measurement differences.

23. Multilingual and Cultural Validity

Language cannot be separated from culture.

Features that appear unusual in one linguistic population may be normal in another.

Important variables include:

primary language;
multilingual status;
code-switching;
dialect;
regional pronunciation;
educational background;
literacy;
genre familiarity;
religious language;
cultural storytelling form;
and disability.

Psychosis-language researchers have specifically warned that multilingualism is insufficiently addressed even though psychiatric assessment depends heavily on language.

A model trained primarily on standardized American English should not be assumed valid for:

Appalachian English;
African American English;
Caribbean dialects;
immigrant populations;
bilingual speakers;
people using augmentative communication;
or people with limited formal education.

Personal baseline can reduce—but not eliminate—this problem.

24. Proposed Clinical Workflow

A practical workflow might proceed as follows.

Step 1: Consent

Explain collection, purpose, retention, access, and limits.

Step 2: Context Capture

Record relevant clinical and environmental conditions.

Step 3: Standardized Prompt

Administer one or more repeatable tasks.

Step 4: Preserve the Original

Store the unaltered writing or audio.

Step 5: Patient Annotation

Ask what the sample means and whether any portion is fictional, metaphorical, or written from another perspective.

Step 6: Automated Feature Extraction

Measure predefined linguistic or acoustic features.

Step 7: Personal Baseline Comparison

Compare with prior samples from the same patient.

Step 8: Clinical Corroboration

Review symptoms, examination, medication, laboratory findings, behavior, and circumstance.

Step 9: Collaborative Interpretation

Discuss significant changes with the patient.

Step 10: Intervention or Observation

Respond proportionately.

Step 11: Outcome Tracking

Determine whether the interpretation and intervention proved useful.

Step 12: Model Correction

Use errors and outcomes to improve—not merely reinforce—the system.

25. Research Program

The framework should be tested prospectively.

Phase I: Feasibility

Determine whether patients will complete standardized expressive tasks and whether collection can be integrated into care without excessive burden.

Phase II: Reliability

Measure whether the same task produces sufficiently stable features under similar conditions.

Phase III: Baseline Construction

Determine how many samples are required to establish a meaningful personal range.

Phase IV: Clinical Association

Test relationships between verbal changes and independently measured clinical outcomes.

Phase V: Longitudinal Prediction

Examine whether changes precede:

symptom worsening;
hospitalization;
cognitive decline;
medication intolerance;
functional loss;
recovery;
or treatment response.

Phase VI: Comparative Utility

Compare verbal telemetry with:

ordinary clinician notes;
symptom questionnaires;
passive smartphone data;
structured interviews;
and standard clinical assessments.

Phase VII: Fairness and Generalizability

Validate across:

age;
sex;
culture;
language;
dialect;
education;
disability;
diagnosis;
and care setting.

Phase VIII: Clinical-Impact Trial

Determine whether using the telemetry actually improves outcomes rather than merely producing interesting predictions.

26. Hypotheses

The following hypotheses are testable.

Hypothesis 1

Changes from personal baseline will be more clinically informative than population-only classification.

Hypothesis 2

Combined linguistic, acoustic, and contextual features will outperform any single feature class.

Hypothesis 3

Structured perspective-shifting tasks will reveal cognitive flexibility not reliably captured by ordinary symptom questionnaires.

Hypothesis 4

Original patient expression plus patient annotation will reduce provider misinterpretation compared with provider summary alone.

Hypothesis 5

Expressive tasks will be acceptable and beneficial to some patients but neutral or distressing to others.

Hypothesis 6

Route constriction measures will correlate with selected clinical outcomes only when objective external constraints and cultural context are included.

Hypothesis 7

Systems that display source evidence and uncertainty will produce safer clinical interpretation than systems providing only a risk score.

Hypothesis 8

Verbal telemetry will have its greatest value as longitudinal decision support rather than stand-alone diagnostic classification.

27. What Success Would Look Like

Success would not be an algorithm that announces a person’s hidden diagnosis after reading a poem.

Success would look like this:

Over the last eight weeks, the patient’s speech rate has slowed, pauses have increased, future-oriented language has declined, and writing has become markedly more repetitive compared with their established baseline. The patient reports worsening pain and poor sleep after a medication change. Clinical review is recommended.

Or:

The patient’s narratives now contain more alternative explanations, greater social reference, and increased future planning than before treatment. These changes align with improved function and self-reported mood.

Or:

The automated system detected an apparent coherence change, but review showed that the patient had switched languages and was completing a creative-fiction task. The alert was dismissed and the model error recorded.

That last example is as important as the first two.

A trustworthy system must learn from being wrong.

28. Limitations

Patient Verbal Telemetry faces major limitations.

Language is not a direct readout of mind.

Creative work may be fictional.

Patients may alter their responses.

Providers may misinterpret samples.

Algorithms may amplify bias.

Recording conditions may vary.

Cultural differences may be mistaken for pathology.

Models may perform well in one clinic and fail elsewhere.

Clinical associations may not produce useful interventions.

Data storage may create privacy risks.

The measurement process itself may change the patient.

Large datasets may increase confidence without increasing truth if the labels, samples, or populations are biased.

More data are useful only when collection, interpretation, validation, and governance improve with them.

29. TSTOEAO Clinical Interpretation

TSTOEAO contributes a way to organize the relationship between expression and available resolution.

The patient is not reduced to words.

Words are treated as evidence of an active system encountering gradients and boundaries.

A repeated narrative may reveal:

which gradients dominate;
which boundaries have become rigid;
which routes remain available;
which routes are suppressed;
where perceived cost is located;
and which resolution repeatedly emerges.

Treatment may then be understood as changing route-space.

Sometimes the relevant intervention is cognitive.

Sometimes it is medication.

Sometimes it is pain treatment.

Sometimes it is housing.

Sometimes it is protection from abuse.

Sometimes it is sleep.

Sometimes it is social connection.

Sometimes it is writing.

Sometimes the patient does not need a new interpretation.

They need a real external boundary changed.

That distinction is critical.

A theory of route-space must never become a method for explaining away material suffering as a defective mindset.

30. Conclusion

Medicine has spent centuries learning to measure the human body.

It has developed instruments for signals too faint, fast, small, internal, or complex for unaided observation.

Yet patient language—one of the most detailed signals the person produces—remains inconsistently collected and frequently reduced to another person’s summary.

The technology now exists to preserve speech, writing, timing, acoustics, semantics, narrative structure, and longitudinal change.

The scientific challenge is not merely to collect more words.

It is to collect them:

consistently;
ethically;
contextually;
transparently;
longitudinally;
and with sufficient humility.

Patient language cannot replace examination, laboratory testing, imaging, clinical interviews, or professional judgment.

It can become another instrument.

The patient’s poem is not a diagnosis.

The patient’s voice is not a verdict.

A repeated change may be a signal.

A recurring boundary may be a question.

A narrowed narrative may identify a place where help is needed.

A reopened route may be evidence of recovery.

The patient has always been speaking.

The patient has always been writing.

The patient has always been transmitting.

The missing instrument is the system capable of listening without pretending that listening makes it infallible.

That instrument is Patient Verbal Telemetry.

References

Tang, S. X., et al. “Automated Speech and Language Markers of Longitudinal Changes in Psychosis Symptoms.” Nature Mental Health, 2025.
Heitz, J., et al. “Towards a Speech-Based Digital Biomarker for Cognitive Functioning.” npj Digital Medicine, 2026.
Emden, D., et al. “Scalable Depression Monitoring with Smartphone Speech.” npj Digital Medicine, 2026.
Da Cunha, E., et al. “Spontaneous Speech Enables Scalable Digital Phenotyping of Multidimensional Physical Function.” npj Aging, 2026.
Berisha, V., et al. “Responsible Development of Clinical Speech AI: Bridging Speech Science and Clinical Research.” npj Digital Medicine, 2024.
Palaniyappan, L. “Could Language Be a Biosocial Marker of Psychosis?” Schizophrenia, 2021.
Baikie, K. A., et al. “Expressive Writing and Positive Writing for Participants with Mood Disorders.” 2012.
Niles, A. N., et al. “Randomized Controlled Trial of Expressive Writing for Psychological and Physical Health.” 2014.
Qian, J., et al. “Effects of Expressive Writing Intervention for Women’s PTSD, Depression, Anxiety and Stress: A Meta-Analysis of Randomized Controlled Trials.” 2020.
Mosher, C. E., et al. “Randomised Trial of Expressive Writing for Distressed Cancer Patients.” 2012.
U.S. Food and Drug Administration. Digital Health Technologies for Remote Data Acquisition in Clinical Investigations. Final guidance.
U.S. Department of Health and Human Services. HIPAA Privacy Rule and Mental Health Information Guidance.
National Institute of Standards and Technology. Artificial Intelligence Risk Management Framework 1.0.
Swygert, John. “Boundary Portfolio Engineering: Opening, Closing, Weighting, Transforming, and Multiplying Physical Route-Space.” The Journal of TSTOEAO, 2026.
Swygert, John. “Light Surfing the Boundary.” The Journal of TSTOEAO, 2026.