A Glossary for Culturally Grounded Behavioral Health

Some of these terms are clinical. Some are technical. Some are policy shorthand that shows up in an RFP or a payer contract without ever being defined. All of them get used elsewhere on this site, usually without a footnote — this page is the footnote. Each definition is written for someone evaluating Vasl, not for someone already fluent in the field, and links to the fuller page where one exists.

AAVE

African American Vernacular English — a rule-governed dialect of English with its own grammar, phonology, and vocabulary, spoken by many Black Americans alongside or instead of Mainstream American English. Standard clinical language models are trained almost entirely on Mainstream American English, so they systematically misread AAVE — not because the dialect is imprecise, but because the model was never taught it.

Behavioral health equity

The absence of avoidable, unfair differences in behavioral health access, quality, and outcomes across groups defined by race, ethnicity, language, or socioeconomic status. In youth mental health, inequity shows up as longer wait times, lower engagement, and higher dropout for Black, Latino, and first-generation populations — not because need is lower, but because the systems built to reach them usually weren't.

Care continuum

The full range of support available to a person, from lowest-intensity (peer community, self-guided check-ins) to highest-intensity (licensed therapy, crisis support), with clear pathways between levels. A continuum only works if movement between levels is easy in both directions — stepping up when needed, stepping down when it's not.

Coded language

Words or phrases a community develops, often deliberately, to communicate sensitive meaning without triggering automated moderation or drawing unwanted attention. “Unaliving” is the most widely cited example. Coded language evolves continuously, which is why a one-time vocabulary update can't keep a detection system current.

Crisis support

The highest-acuity layer of a behavioral health system, activated when a member may be at imminent risk. Vasl surfaces signals that may warrant crisis support to a human — a coach, a counselor, a clinician — who makes the actual determination and initiates response. The system does not make crisis determinations on its own.

Cultural humility

An ongoing practice of self-reflection about one's own cultural assumptions, paired with genuine curiosity about someone else's, rather than a credential earned once. In clinical training it's often contrasted with “cultural competence,” which can imply a fixed body of knowledge is sufficient.

Cultural idioms of distress

Culturally specific ways of expressing or experiencing psychological pain that don't map cleanly onto standard clinical categories — a phrase, a somatic complaint, a way of describing a feeling that carries meaning inside a community but reads as ambiguous or even invisible outside it. Reading these correctly requires a model to have learned them, not translate them from a standard baseline.

Culturally responsive care

Support designed around a population's language, communication norms, and lived context from the start, rather than a standard model retrofitted with cultural add-ons afterward. The distinction matters because retrofitting tends to preserve the original model's blind spots.

Dialect bias

Systematic misreading of meaning, tone, or severity by a system trained primarily on one dialect when it encounters another. In clinical NLP, dialect bias can cause identical distress language to be scored differently depending on which dialect it's expressed in. See differential item functioning.

Differential item functioning

A measurement phenomenon where a test item behaves differently across groups even when the underlying trait it measures is held equal — meaning an identical score does not always reflect an identical clinical reality. Differential item functioning has been documented in standard screening instruments including the PHQ-9 across race and ethnicity. Read the full explainer.

Engagement

Active, ongoing use of a behavioral health platform or service, as distinct from enrollment or access alone. A member can be enrolled and never meaningfully engage; engagement is the metric that actually predicts benefit. Vasl reports 30-day retention and time to first substantive interaction as engagement indicators.

Escalation pathway

The defined route by which a concern moves from a lower-intensity layer of support to a higher one — for example, from a peer community moderator to a coach, or from a coach to a licensed clinician — with clear criteria for when and how the handoff happens.

False negative

An instance where a detection system fails to flag a genuine signal of distress. In screening and detection contexts, false negatives are generally treated as the more serious error type, since they mean a person who needed support wasn't identified. Sensitivity is the metric most directly tied to minimizing false negatives.

FERPA

The Family Educational Rights and Privacy Act — the federal law governing the privacy of student education records. School-based behavioral health tools need to be architected around FERPA's boundaries on what counts as an education record and who can access it, distinct from HIPAA, which governs health records in clinical settings.

FQHC

A Federally Qualified Health Center — a community-based clinic that receives federal funding to provide primary and preventive care, including behavioral health, in underserved areas, regardless of a patient's ability to pay. FQHCs are a common deployment setting for youth behavioral health programs serving Medicaid populations.

FUH

Follow-Up After Hospitalization for Mental Illness — a HEDIS measure tracking whether a patient received outpatient follow-up care within 7 and 30 days of a mental health hospitalization. FUH is considered a strong proxy for care coordination quality, and follow-up completion depends heavily on whether the patient stays engaged after discharge.

FUM

Follow-Up After Emergency Department Visit for Mental Illness — a HEDIS measure similar to FUH, but tracking follow-up after an emergency department visit rather than an inpatient stay. Like FUH, FUM outcomes depend heavily on whether a patient can be reached and re-engaged quickly after the ED encounter.

Help-seeking mismatch

The gap between how a person actually asks for help — often indirectly, in coded or culturally specific language — and how a system is built to recognize a request for help. When the mismatch is large, real requests go unanswered not because no one was listening, but because the system was listening for the wrong signal.

HEDIS

The Healthcare Effectiveness Data and Information Set — a standardized set of performance measures used by health plans to report on care quality, including several specific to behavioral health engagement and follow-up. Read the full breakdown.

HIPAA

The Health Insurance Portability and Accountability Act — the federal law governing the privacy and security of individually identifiable health information in clinical and payer settings. Vasl's architecture is built around HIPAA's boundaries on protected health information, distinct from the FERPA boundaries that apply in school settings.

Human-in-the-loop

A system design principle where a human reviews and makes the final determination on any consequential output, rather than the system acting autonomously. Vasl surfaces signals to a human — never to an automated decision or an external authority — who applies clinical and relational judgment before anything happens.

IET

Initiation and Engagement of Substance Use Disorder Treatment — a HEDIS measure tracking whether a patient with a new SUD diagnosis both initiates treatment and stays engaged with it over time. IET is a two-part measure precisely because starting treatment and remaining in it are different challenges with different drop-off points.

Large language model

A machine learning model trained on large volumes of text to represent and generate language. General-purpose large language models are trained predominantly on internet text that underrepresents culturally specific and dialect-marked language — the starting problem VLAP was built to address.

Linguistic masking

Expressing distress in a form that minimizes social risk or the chance of an unwanted response — hedged, indirect, or coded language that carries real weight to someone who reads it correctly, but reads as minor or ambiguous otherwise.

MTSS

Multi-Tiered System of Supports — a framework used in K-12 education for matching the intensity of support to the intensity of student need, typically organized in three tiers. See Tier 1/2/3 supports.

Natural language processing

The field of computing concerned with how systems process, interpret, and generate human language. Clinical NLP applies these methods to language generated in health and behavioral health contexts, where the cost of misreading meaning is higher than in most other NLP applications.

Peer support specialist

A trained facilitator, often with lived experience relevant to the population served, who supports peer community spaces and provides connection that does not require licensure. Peer support sits at the access layer of a care continuum — the relationship that exists before a clinical need is identified.

PHQ-8

The Patient Health Questionnaire, 8-item version — a self-report depression screening instrument, functionally identical to the PHQ-9 minus the item on suicidal ideation. Vasl uses the PHQ-8 as its primary outcome instrument for pilot cohort reporting because it is a widely validated short-form screener suitable for non-clinical administration.

PHQ-9

The Patient Health Questionnaire, 9-item version — the PHQ depression screener including an item on thoughts of self-harm. Research has documented differential item functioning in the PHQ-9 across race and ethnicity, meaning the same score does not always represent the same underlying symptom severity across groups.

Population-level insight

Aggregate, de-identified patterns across a group of members — trends, not individual records. Vasl's organizational reporting operates at this level; a minimum cohort size is enforced before any aggregate data is surfaced, specifically to prevent identifying an individual by inference from a small group.

Protected health information

Individually identifiable health information subject to HIPAA's privacy and security requirements. Vasl's platform reporting to organizational administrators is built to exclude protected health information — no member name, session content, or individual clinical detail is accessible outside the direct care relationship.

Retention

The proportion of enrolled members who remain actively engaged with a platform or service over a defined period. Vasl reports 30-day retention because early disengagement is the point at which most behavioral health interventions lose their ability to help.

Sensitivity

In detection and screening contexts, the proportion of true positive cases a system correctly identifies — the inverse of the false negative rate. High sensitivity is prioritized in early-detection contexts because the cost of missing a genuine signal generally outweighs the cost of a false alarm reviewed and dismissed by a human.

Teletherapy

Licensed mental health therapy delivered remotely, typically by video or phone, by a clinician credentialed in the client's state. Teletherapy extends the reach of licensed care without requiring a young person to access it in person, which matters most in areas with clinician shortages.

Tier 1/2/3 supports

The three levels of MTSS: Tier 1 is universal support available to every student; Tier 2 is targeted support for students showing early signs of need; Tier 3 is intensive, individualized support for students with significant need. A well-mapped program has clear criteria for moving between tiers in both directions.

Transition-age youth

Young people roughly ages 16–25 moving out of child-serving systems (pediatric care, K-12 schools, foster care) and into adult-serving ones — a period associated with significant drop-off in behavioral health engagement, since eligibility, providers, and sometimes insurance all change at once.

Universal screening

Administering a standardized assessment to an entire population, rather than only to individuals already flagged for concern, so that need is identified consistently rather than depending on who happens to ask for help or who happens to be noticed.

Warm handoff

A transfer of care between two people — for example, a coach and a licensed clinician — that includes direct introduction and context-sharing, rather than a referral the member has to navigate alone. Warm handoffs are associated with meaningfully higher follow-through than cold referrals.