Here’s a glossary of common terms in Conversational AI:
This is for those who aren't tech savvy and would like to understand conversational AI in easier terms.
A
-
Agentic AI – Like a super-smart assistant that can set its own goals, make plans, and figure out how to get things done—without constant instructions.
-
API (Application Programming Interface) – Think of it as a translator that helps different software programs talk to each other.
-
ASR (Automatic Speech Recognition) – Turns what you say out loud into written text so a computer can understand it.
-
Autonomous Agent – A digital helper that can make its own decisions and take action with little or no human input.
B
-
Bias – When an AI makes unfair decisions or assumptions because it learned from flawed or unbalanced data.
-
Bot – A software program (like a chatbot or voice assistant) that talks with people automatically.
-
BM25 – A smart way of ranking search results based on how well they match your keywords.
C
-
Chatbot – A virtual assistant you can text with, often found on websites or customer service apps.
-
Chunking – Cutting big documents into smaller pieces so they’re easier for AI to search through.
-
Conversational AI – Tech that lets machines talk with you in a natural, human-like way.
-
Context Window – The amount of conversation the AI can “remember” at one time.
D
-
Dialogue Management – The behind-the-scenes rules that help an AI keep a conversation flowing smoothly.
-
Domain-Specific – Designed to work really well in a certain field, like healthcare or law.
E
- Embedding – A way of turning words into numbers so the AI can understand what they mean.
- Entity Recognition (NER) – When an AI picks out important bits like names, dates, or places in a sentence.
- Episodic Memory (in agents) – The AI’s ability to remember past chats or events to stay consistent.
F
-
Few-shot Learning – Teaching an AI a new task using just a few examples.
-
Fine-tuning – Customising an AI so it’s better at a specific job by training it on new data.
-
Feedback Loop – A learning system where the AI gets better by adjusting based on what worked or didn’t.
G
-
Goal-Oriented Agent – An AI focused on achieving a specific result, like booking a flight or solving a problem.
-
Grounding – Making sure the AI gives answers based on real, reliable info.
H
-
Hallucination – When an AI makes something up that sounds real but isn’t.
-
Hybrid Search – Mixing two search styles—by meaning and by keyword—for better results.
I
-
Intent Recognition – Figuring out what someone really wants when they say or type something.
-
Inference – When the AI thinks through a question and comes up with an answer.
J
-
JSON – A simple format for organising data so different apps can share it easily.
K
-
Knowledge Base – A digital library of facts and documents that an AI can use to answer questions.
-
Knowledge Grounding – Making sure the AI is using solid sources to back up what it says.
L
-
LLM (Large Language Model) – A very powerful AI trained on tons of text to understand and generate language.
-
LangChain – A toolkit that helps developers build apps that combine AI with search, tools, and memory.
M
-
Memory (Agent Memory) – The AI’s ability to remember what happened in past conversations or tasks.
-
Multimodal AI – AI that can handle text, speech, images, or videos—kind of like how humans use different senses.
N
-
Natural Language Processing (NLP) – The part of AI that helps it understand and use human language.
-
NLG (Natural Language Generation) – The AI writing or saying something in a way that sounds natural.
-
NLU (Natural Language Understanding) – The AI figuring out what you really meant when you said something.
O
-
Open-Domain Chatbot – A chatbot that can talk about almost anything, not just a narrow topic.
-
Ontology – A structured map of knowledge that helps AI organise and reason about information.
P
-
Prompt Engineering – Carefully crafting questions or commands to get the best results from AI.
-
Planning (in Agentic AI) – When an AI thinks through the steps needed to get something done.
-
Post-Retrieval Reranking – Improving search results by re-checking them for quality.
Q
-
Query Expansion – Adding related words to a search to help the AI find better results.
R
-
RAG (Retrieval-Augmented Generation) – A smart combo where the AI looks things up and uses its own knowledge to answer.
-
Retriever – The part of the system that finds documents relevant to your question.
-
Reinforcement Learning (RLHF) – Teaching the AI through trial, error, and feedback—like how kids learn.
-
Reflection (in agents) – The AI’s ability to look back at what it did and learn from it.
S
-
Semantic Search – Searching by meaning, not just exact words.
-
Slot Filling – Collecting all the info needed to complete a task, like filling out a form during a chat.
-
Speech Synthesis (TTS) – Turning written text into spoken words using a computer voice.
T
-
Task Decomposition – Breaking a big job into smaller steps so it’s easier to manage.
-
Tool Use (in Agentic AI) – When the AI uses external tools (like search engines or calculators) to help solve a problem.
-
Transformer – The special brain-like structure inside most modern AI's that helps them understand language.
U
- Utterance – One thing someone says in a conversation, like a question or a reply.
- User Simulation – Fake user conversations used to train or test AI systems.
V
-
Vector Database – A place to store the AI’s understanding of concepts (as numbers) so it can search by meaning.
-
Voicebot – A chatbot you can speak to instead of typing, like Alexa or Siri.
W
-
Workflow Automation (Agentic AI) – Letting AI take over long, multi-step tasks so people don’t have to.
X
-
Explainability (XAI) – Helping people understand how and why AI made a decision.
Y
-
Yield Optimisation (Conversational) – Making sure AI conversations lead to better results, like more sales or happier customers.
Z
-
Zero-shot Learning – When AI can do something it hasn’t seen before just by using what it already knows.
