Glossary
SLPs know 'pragmatics' but not 'temperature.' Developers know 'hallucination' but not 'functional communication.' Each term is defined from both directions.
Hallucination
Clinical: When an LLM generates information that sounds authoritative but is factually incorrect, like a confident student who gue...
Tech: A model output that is not grounded in the training data or input prompt, produced because the model generates statistic...
Pragmatics
Clinical: The rules governing how language is used in social contexts: turn-taking, topic maintenance, inferencing, understanding ...
Tech: In AI/NLP, pragmatics relates to how context influences meaning beyond the literal words, including understanding implic...
Temperature
Clinical: A setting in AI tools that controls how predictable or creative the output is. Lower temperature (0.0-0.3) produces more...
Tech: A parameter in language model inference that scales the probability distribution over the vocabulary before sampling. Te...
Prompt
Clinical: The text you type into an AI tool to tell it what you want. Think of it like a referral question: the more specific and ...
Tech: The input text sequence that conditions a language model's output distribution. The prompt provides the context from whi...
Token
Clinical: The smallest unit of text that an LLM processes. Tokens are like morphemes, the smallest meaningful pieces the model wor...
Tech: A subword unit produced by a tokenization algorithm (such as BPE or SentencePiece) that splits text into a vocabulary of...
Context Window
Clinical: The model's working memory: the total amount of text it can 'hold in mind' during a conversation. Think of it like audit...
Tech: The fixed-length sequence of tokens that a transformer model can attend to during inference. All input tokens (system pr...
Fine-Tuning
Clinical: The process of taking a general-purpose AI model and training it further on specialized data to make it better at a spec...
Tech: A transfer learning technique in which a pre-trained language model undergoes additional training on a smaller, domain-s...
PHI (Protected Health Information)
Clinical: Any individually identifiable health information that is created, received, maintained, or transmitted by a healthcare p...
Tech: A category of data defined under the HIPAA Privacy Rule (45 CFR 160.103) encompassing 18 specific identifiers linked to ...
BAA (Business Associate Agreement)
Clinical: The legal contract between you (or your employer) and a third-party service provider that establishes how they will prot...
Tech: A contract required under HIPAA (45 CFR 164.502(e)) between a covered entity and a business associate that handles prote...
Scope of Practice
Clinical: The ASHA-defined boundaries of what speech-language pathologists are qualified and authorized to do, including assessmen...
Tech: Professional domain constraints that define the boundaries within which AI tools should operate when used by a specific ...
Model
Clinical: A trained AI system that generates text by predicting what comes next. Think of it like a clinician's accumulated knowle...
Tech: A neural network trained on large volumes of text data to predict the probability of the next token in a sequence. Model...
Grounding
Clinical: Connecting AI output to specific source material rather than letting the model generate from its training data alone. Th...
Tech: A technique for constraining model output to information contained in provided reference documents, typically implemente...
De-identification
Clinical: Removing all identifying information from clinical data before it enters any AI tool. This is not optional. Under HIPAA,...
Tech: The process of removing or transforming personally identifiable information (PII) and protected health information (PHI)...
System Prompt
Clinical: Hidden instructions that shape how an AI tool behaves before you ever interact with it. Like an IEP that governs a thera...
Tech: The initial instruction set in a conversational AI message sequence, typically invisible to the end user, that establish...
Guardrails
Clinical: Safety constraints that prevent an AI tool from doing things it shouldn't, like scope of practice boundaries for softwar...
Tech: A combination of content filtering, output validation, safety-trained model behavior, and application-level restrictions...
RAG (Retrieval-Augmented Generation)
Clinical: Instead of the model guessing from memory, it looks up specific documents first, then generates a response based on what...
Tech: An architecture that combines information retrieval with language model generation. A user query is first used to search...
Evidence-Based Practice
Clinical: The clinical decision-making framework defined by ASHA as the integration of three components: best available research e...
Tech: A decision-making framework originating in healthcare that requires integrating empirical evidence, practitioner experti...
Functional Communication
Clinical: Communication that works in real life: not perfect articulation in a therapy room, but the ability to get needs met, bui...
Tech: A concept that AI tools have no mechanism to assess. LLMs can generate grammatically correct, contextually appropriate l...
Clinical Voice
Clinical: The individual perspective a clinician brings to documentation: your specific observations, your professional phrasing, ...
Tech: The stylistic and semantic properties of text that reflect an individual author's perspective, expertise, and reasoning ...
Informed Consent
Clinical: The ethical obligation to inform clients and families when AI tools are used in any part of their care, including docume...
Tech: Transparency requirements around AI use in clinical settings. As AI tools become embedded in healthcare documentation wo...
Measurability
Clinical: Whether a goal can actually be tracked with data. This is the make-or-break quality of any IEP or treatment goal. A meas...
Tech: The operationalizability of a stated objective: whether it can be converted into a concrete, observable, and quantifiabl...
Copilot
Clinical: The recommended mental model for using LLMs in clinical work: the AI organizes, structures, and drafts, but you make eve...
Tech: A human-in-the-loop AI assistance paradigm where the model's role is constrained to support tasks while a human retains ...
Bias
Clinical: Systematic patterns in AI output that reflect skewed assumptions from training data. In SLP contexts, this shows up as d...
Tech: Systematic deviations in model output that reflect the distributions, perspectives, and gaps in training data. Bias in L...
Interoperability
Clinical: Whether your AI tool can connect with your EMR, IEP software, or billing system, and the privacy implications when it do...
Tech: The ability of different software systems to exchange, interpret, and use data across boundaries without manual interven...
Natural Language Processing
Clinical: The technology underlying every AI tool SLPs encounter, and the irony is thick. SLPs are the clinical experts in human l...
Tech: The subfield of artificial intelligence and computer science focused on enabling computers to process, analyze, and gene...