Glossary

Search results

Automated Speech Recognition(also: ASR, Speech-to-Text, Voice Recognition)

Technology that converts spoken language into written text using machine learning and signal processing algorithms. In accessibility, ASR is used for real-time captioning, voice control of devices and software, and generating transcripts of audio and video content. While ASR…

Automatic Alt Text(also: AI-generated Alt Text, Auto Alt Text, Machine-generated Alt Text)

Alternative text for images that is automatically generated by artificial intelligence systems using computer vision and natural language processing. Platforms like Microsoft PowerPoint, Facebook, and Twitter have incorporated automatic alt text features to increase image…

Automatic Image Captioning(also: AI-Generated Alt Text, Machine-Generated Description, Auto Alt Text)

The use of computer vision and machine learning algorithms to automatically generate textual descriptions of images without human intervention. Social media platforms like Facebook and Instagram deploy automatic image captioning to provide alternative text for user-uploaded…

Automatic Speech Recognition(also: ASR)

The computational technology and algorithms that automatically convert spoken audio into text transcriptions. ASR systems use acoustic models, language models, and increasingly deep learning approaches to process speech signals. ASR accuracy directly impacts the usability of…

BERT(also: Bidirectional Encoder Representations from Transformers)

A natural language processing model developed by Google that uses bidirectional training to understand context from both directions in a sentence. BERT and its variants like SBERT (Sentence-BERT) are increasingly used in accessibility applications for tasks such as automatic…

Be My AI

An AI-powered feature within the Be My Eyes app that uses GPT-4o to provide on-demand image descriptions for blind and low vision users. Users can take a photo or upload an image and receive a detailed AI-generated description, replacing the need to connect with a sighted…

Benchmark dataset(also: Evaluation dataset, Test benchmark)

A standardized dataset used to evaluate and compare the performance of AI models, algorithms, or systems against established baselines. In accessibility, the absence of benchmark datasets that include people with disabilities means disparate performance across disability…

Bi-Directional Alignment(also: Two-Way Alignment, Mutual Alignment)

An approach to human-AI interaction design that addresses alignment from both directions: not only adapting AI systems to match human values and preferences, but also helping humans understand, direct, and correct AI system behaviour. Traditional AI alignment focuses solely on…

Bias Benchmarking Questionnaire(also: BBQ)

A standardized dataset used in AI fairness research to evaluate social biases in language models. The BBQ consists of carefully crafted context-question pairs designed to test whether models exhibit stereotypical associations related to age, gender, race, disability, and other…

Bias Mitigation(also: Algorithmic Fairness, Debiasing)

The process of identifying and reducing systematic errors or prejudices in AI systems, datasets, and algorithms that lead to unfair outcomes for particular groups of people. In accessibility, bias mitigation is critical because AI training datasets often underrepresent people…

Biometric System(also: Biometric Technology, Biometric Identification)

A technology system that uses innate human physical or behavioral characteristics — such as facial features, fingerprints, voice patterns, gait, or iris patterns — to identify or verify a person's identity. Biometric systems pose particular risks for people with disabilities…

Black Box Model(also: Opaque Model)

A machine-learning model whose internal workings are not directly inspectable or interpretable by a human, either because the model is architecturally complex (deep neural networks, large language models) or because it is proprietary and the developer does not disclose its…

CLIP(also: Contrastive Language-Image Pre-Training)

A vision-language model developed by OpenAI that learns to associate images with natural language descriptions through contrastive learning on large-scale image-text pairs. CLIP can compute similarity scores between images and text, enabling zero-shot classification and…

Cascading classifier(also: Staged classifier, Multi-stage classifier)

A machine learning architecture that applies progressively more computationally expensive analysis stages, with each stage filtering out easy-to-classify cases so that only ambiguous instances proceed to deeper analysis. In accessibility applications, cascading classifiers…

Chain-of-Thought Prompting(also: CoT Prompting)

A technique for improving the reasoning capabilities of large language models by instructing them to break down complex tasks into intermediate reasoning steps before producing a final answer. In accessibility applications, chain-of-thought prompting is used to improve the…

ChatGPT Accessibility(also: GenAI Accessibility)

The degree to which ChatGPT and similar generative AI interfaces can be effectively used by people with disabilities, including compatibility with screen readers, keyboard navigation, and non-visual interaction patterns. Research with teachers with vision impairments in India…

Chatbot(also: Virtual Assistant, Conversational Agent, Dialog System)

A software application that uses text or speech to conduct conversations with users, simulating human-like dialogue to provide information, perform tasks, or guide users through processes. In accessibility, chatbots and conversational user interfaces present both opportunities…

Circumplex Model of Affect(also: Russell's Circumplex Model, Valence-Arousal Model)

A psychological framework that represents emotions along two continuous dimensions: valence (pleasure vs. displeasure) and arousal (activation vs. deactivation). Proposed by James Russell in 1980, the model maps all emotional states onto a circular space rather than treating…

Clue and Reasoning Prompting(also: CARP, Clue-and-Reasoning Prompting)

A prompt engineering strategy for large language models that instructs the model to first identify textual clues (keywords, phrases, contextual information) in the input and then perform diagnostic reasoning based on those clues before producing a classification output.…

Cognitive Assistance(also: Cognitive Aid, AI-Powered Assistance, Assisted Cognition)

Technology that uses artificial intelligence and machine learning to supplement or expand human cognitive and perceptual abilities. In accessibility contexts, cognitive assistance systems recognise people, objects, text, and environments and convey that information through…

Computer Vision(also: CV, Machine Vision)

A field of artificial intelligence that enables computers to interpret, understand, and extract information from visual data including images and videos. Computer vision technologies—such as object detection, image segmentation, scene recognition, and optical character…

Computer-Using Agent(also: CUA)

An AI agent, typically built on a Large Multimodal Model, that perceives a computer's graphical user interface through screenshots, reasons about on-screen context, and directly manipulates the interface by clicking, typing, scrolling, and navigating between applications. Unlike…

Confidence Score(also: Confidence Rating, Certainty Score)

A numerical measure output by AI systems indicating how certain the system is about a particular result or classification. While confidence scores have been proposed as a way to help users assess AI accuracy, research with blind participants has found them difficult to interpret…

Constitutional AI(also: CAI)

A training method introduced by Anthropic in 2022 in which a large language model is aligned to a written set of principles (a 'constitution') through self-critique and reinforcement learning from AI feedback, rather than relying exclusively on human preference labels. The model…

Content Moderation(also: Content Filtering, Automated Content Moderation)

The process of monitoring and filtering user-generated content on digital platforms, increasingly performed by AI systems. Content moderation has documented negative effects on people with disabilities: automated systems have suppressed content from disabled creators (TikTok…

Context Retention(also: Conversational Context, Context Awareness)

The ability of a voice assistant or AI system to maintain awareness of previous interactions and use that information to interpret subsequent commands correctly. In calendar accessibility, context retention is important because scheduling tasks often involve multi-turn…

Contextual Learning(also: Context-Dependent Learning)

The tendency of both humans and AI systems to learn patterns and behaviours from the surrounding context rather than from abstract rules. In web development, contextual learning means that developers working on accessible codebases are more likely to produce accessible code…

Continual Learning(also: Continuous Learning, Lifelong Learning, Never-ending Learning)

A machine learning paradigm in which models learn incrementally from new data over time while retaining previously acquired knowledge, rather than being trained once on a fixed dataset. Continual learning is relevant to accessibility because it enables AI-powered accessibility…

Conversational AI(also: Chat AI, AI Chatbot)

Artificial intelligence systems designed to engage in dialogue with users through natural language, including chatbots, virtual assistants, and generative AI interfaces. Conversational AI has accessibility implications both as an interaction paradigm that can be more accessible…

Conversational Agent(also: Chatbot, Virtual Assistant, AI Assistant)

A software system that uses natural language processing to engage in dialogue with users, answering questions and providing information through text or speech. In accessibility contexts, conversational agents offer potential for flexible, on-demand information access that can…

Conversational Assistant(also: Voice Assistant, Virtual Assistant, Intelligent Personal Assistant)

A software application that uses natural language processing and speech recognition to interact with users through spoken or typed conversation, providing information, performing tasks, and controlling devices. Examples include Google Assistant, Amazon Alexa, Apple Siri, and…

Convolutional Neural Network(also: CNN, ConvNet)

A class of deep neural network that uses convolutional filters to automatically extract spatial features from data, originally designed for image processing but now widely applied to sensor data, audio, and video analysis. CNNs identify patterns like edges, textures, and shapes…

Counterfactual Explanation(also: Counterfactual XAI)

An explanation technique that communicates what minimal change to the input would have produced a different output from an AI model, for example 'if the applicant's income had been $5,000 higher, the loan would have been approved'. Counterfactual explanations are legally…

Counterfactual Prompting(also: Counterfactual Debiasing, Counterfactual Data Augmentation)

A bias mitigation technique that involves modifying prompts or training examples by swapping identity-related attributes (such as disability status, gender, or race) while keeping all other context identical, in order to expose and counteract biased associations in language…

Cross-Checking(also: Cross-Verification, Multi-Tool Verification)

A verification strategy used by blind and low vision people to assess the reliability of AI-generated image descriptions by comparing outputs from multiple AI tools, taking photos from different angles, using non-visual senses, or consulting sighted individuals. BLV users have…

Crowd Accessibility(also: Crowdsourcing for Accessibility, Human-Powered Access Technology)

An approach that combines human intelligence with machine intelligence to create accessible content and services for people with disabilities. In crowd accessibility, micro-tasks that automated systems cannot yet perform reliably — such as describing images, identifying objects,…

Crowd-AI System(also: Hybrid Crowd-AI, Human-AI System)

A system that combines human crowdsourced input with artificial intelligence to accomplish tasks that neither can handle well alone. In accessibility contexts, crowd-AI systems are used for visual question answering, image description, and environmental sensing. Crowd workers…

Datasheets for datasets(also: Dataset documentation, Data cards)

A standardized documentation framework proposed by Gebru et al. that accompanies machine learning datasets with information about their creation, composition, intended use, and limitations. For accessibility, datasheets help surface representation gaps — such as whether people…

Decision-Theoretic Planning(also: Decision-Theoretic Approach)

A computational approach to planning that uses mathematical models of decision-making under uncertainty, most commonly Markov Decision Processes (MDPs), to select optimal actions based on the current state of the system and predicted outcomes. In assistive technology,…

Deep Learning(also: DL)

Deep learning is a subset of machine learning that uses artificial neural networks with multiple layers to learn complex patterns from large amounts of data. In accessibility contexts, deep learning powers many assistive technologies including automatic image description for…

Deepfake(also: Synthetic Media, AI-Generated Media)

AI-generated or AI-manipulated media (images, video, audio, or text) designed to convincingly depict events, people, or statements that never occurred. Deepfakes pose specific risks for people with disabilities: AI-generated fake images of disabled people have been used for…

Description Logic(also: DL, Description Logics)

A family of formal knowledge representation languages used as the mathematical foundation for ontology languages like OWL. Description logics describe a domain in terms of individuals (specific entities), concepts or classes (sets of individuals with shared characteristics), and…

Differential privacy(also: DP)

A mathematical framework for sharing statistical information about a dataset while providing provable guarantees that individual records cannot be identified. In accessibility contexts, differential privacy is proposed as a way to resolve the tension between collecting…

Digital mental health(also: E-mental health, Digital therapeutics)

The use of digital technologies including AI chatbots, mood tracking apps, and online therapeutic platforms to support mental health assessment, monitoring, and intervention. Designing accessible and explainable AI for digital mental health is particularly challenging because…

Directional Stimulus Prompting(also: DSP)

A prompt engineering technique for large language models that provides specific keywords or directional stimuli to guide the model toward generating output focused on particular aspects or attributes. In accessibility applications, DSP is used to produce targeted,…

Disability-Centered Dataset(also: Disability-First Dataset, Accessibility Dataset)

A research dataset specifically designed to capture the practices, environments, and experiences of people with disabilities, rather than retrofitting general-purpose datasets for accessibility evaluation. Disability-centered datasets reflect the real variability, messiness, and…

Domain Adaptation(also: Cross-Domain Transfer, UDA, Unsupervised Domain Adaptation)

A machine learning technique that enables models trained on data from one domain (such as web interfaces) to perform well on a different but related domain (such as mobile app interfaces). Domain adaptation is valuable for accessibility because it allows models trained on…

Dynamic Bayesian Network(also: DBN, Temporal Bayesian Network)

A probabilistic graphical model that represents sequences of variables over time, extending standard Bayesian networks to handle temporal relationships. In accessibility and affective computing contexts, Dynamic Bayesian Networks are used to model how facial expressions, head…

Dynamic Content Filtering(also: Intelligent Content Filtering, AI-Based Filtering)

The automated process of selectively showing or hiding web content based on computed relevance to a user's goals, preferences, or context. Dynamic content filtering uses AI models (such as large language models) to assess which page elements are relevant to a specific task and…

Edge Computing(also: Edge AI, Edge Intelligence)

Edge computing is a computing paradigm where data processing occurs on devices physically close to the user rather than in centralized cloud servers. For accessibility applications, edge computing offers important advantages including reduced latency for real-time assistive…