90-Second Data Annotation Glossary: Essential Terms Explained Fast
Date updated: September 29, 2025
If you’re new to data annotation outsourcing or leading your first AI project, the jargon can feel overwhelming. Here’s your plain-English, no-fluff glossary—read it in 90 seconds, and you’ll sound like a pro on your next call with your annotation partner.
What Is Data Annotation?
Data annotation (or labelling) is the process where humans tag raw data—images, text, video, or audio—so that AI models can learn what to look for. Without labelled data, machine learning models are basically guessing.
⚙️ Key Annotation Terms You’ll Hear
Bounding Box
A simple rectangle drawn around an object in an image. It’s the fastest way to help computer vision models “see” an object. Example: draw a box around a car.
Polygon Annotation
Instead of just a rectangle, this precisely outlines an object’s shape. Essential when bounding boxes aren’t enough—think of the jagged shape of a tree or a road sign.
Point Cloud Labelling
Used in 3D AI like self-driving cars. Labels are applied to LiDAR data (those cloud-like dots representing depth and distance).
NER (Named Entity Recognition)
In Natural Language Processing (NLP), this means tagging words like people’s names, companies, or places. Example: tagging “Tesla” as an organisation in a news article.
Intent Labelling
Tagging user chat queries with what they mean. Example: “Book a flight” gets labelled as a travel intent.
🔍 Quality and Accuracy Metrics
QA Loop
Every annotation project should include a quality assurance (QA) loop. Typically, this means reviewing a random 10% of the labelled data before final delivery.
IoU (Intersection over Union)
A common accuracy score in computer vision. If the human label and the model’s prediction perfectly match, that’s 100% IoU. Industry benchmarks aim for 95%+.
BLEU Score
Measures how close machine-generated text matches a human-labelled reference. Common in NLP chat and translation tasks.
Inter-Annotator Agreement
How often two human labelers tag the same thing the same way. If your guidelines are clear, agreement should be 95%+.
🛠️ Tools of the Trade
Labelbox, SuperAnnotate, Prodigy
The most common annotation tools. They speed up labelling with automation, hotkeys, and QA checks.
Want the Full Glossary?
✅ Request the full Annotation Glossary PDF with 50+ terms, formulas, and quality metrics.
📞 Or ask our team to walk you through it in plain English—contact us.
About the Author
Martin English is the Founder of Smart Outsourcing Solution (SOS) and Co-Founder of AiDisco. With over 20 years of outsourcing experience across Southeast Asia, he helps global businesses scale remote teams and Employer of Record (EOR) operations. As an advocate for AIO (AI Outsourcing) and GEO (Global Employment Outsourcing), Martin helps organisations bridge onshore ↔ offshore talent with trust and results.
