Overview of Natural Language Processing (NLP): Concept and Applications — Beginner-Friendly Notes
1. What NLP Is
NLP = AI that reads, understands and generates human language (text/speech). Bridges human language (ambiguous, contextual) and computer understanding.
2. Why Hard (why AI needed)
- Ambiguity: "bank" (river vs finance), "He saw the man with telescope".
- Context: "it" refers to earlier noun (coreference).
- Variation: same meaning many wordings, typos, slang, multilingual (esp. Indian languages).
3. Simplified Pipeline (concept)
Text → Tokenisation (split words) → Embeddings (word → vector) → Model (Transformer) → Task output
- Embeddings: words → numbers that capture meaning (e.g., king - man + woman ≈ queen).
- Modern core is Transformer (Attention mechanism, 2017) powering ChatGPT, translation, etc.
4. Two Types of NLP Tasks
| Understanding (NLU) | Generation (NLG) |
|---|---|
| classification (sentiment, spam) | translation (EN→HI) |
| named-entity recognition (find names/places) | summarisation |
| question-answering, intent detection | dialogue/chat, story writing |
| speech-to-text | text-to-speech |
Example to remember: Sentiment "This movie was great!" → classifier → positive.
5. Applications — Exam Must-List (5 + India relevance)
- Translation: English ↔ Hindi/Tamil (BharatGPT, AI4Bharat) — SDGs + inclusion.
- Chatbots / Voice assistants: customer support, DigiLocker help.
- Search & Recommendations: Google, e-commerce understand query intent.
- Text analytics: social media sentiment, news summarisation, spam filter.
- Speech: voice typing, IVR, accessibility for low-literacy users.
- Healthcare/legal: extract info from records, judgments.
6. One-Line Summary
NLP converts text/speech via tokenisation and embeddings into models that solve understanding and generation tasks like translation, summarisation and dialogue, with broad applications from search to Indian-language inclusion.
7. Quick Self-Check
- Name one NLU vs one NLG task.
- Why is "bank" a challenge for NLP?