2. Topic 5: Overview
• Introduction
• NLP in Action
• Generative AI and Textual Interfaces
• LLMs; impact, bene
fi
ts and future
2
3. • The Natural Language Interface as a user interface
where linguistic phenomena such as verbs,
clauses, phrases act as controls for creating,
selecting and modifying data in applications.
4. Techniques and
Technologies
• Tokenisation and Text Normalisation
• Part of Speech Tagging (POS) and Named Entity
Recognition (NER)
• Dependency Parsing
• Sentiment Analysis
• Machine Translation
7. How do NLP models help improve
accuracy of textual interfaces?
• Understanding Context
• Transfer Learning
• Handling Ambiguity
• Entity Recognition and Classi
fi
cation
• Sentiment Analysis
• Continuous Learning and Adaptation
• Integration with Domain Speci
fi
c Knowledge
• Addressing Data Ambiguities
• Predictive Text and Autocorrection
8. Limitations of NLP Models
• Ambiguity in Language
• Sarcasm and Irony Detection
• Handling Idiomatic Expressions
• Adapting to Language Evolution
• Data Ambiguities
• Information Overload
• Domain-speci
fi
c Language
• Tokenisation Challenges
• Context Dependency
9. Class Activity
Divide into small groups (ideally 3-4 students each).
Each group takes one text sample
Analyse the text sample assigned to you
Discuss the following questions:
• What is the main topic of the text?
• What are some key entities or phrases mentioned in the text
(e.g., people, organisations, locations)?
• What is the overall sentiment of the text (positive, negative,
neutral)?
10. Generative AI
• Generative AI refers to a class of artificial
intelligence that specialises in creating content,
which can include text, images, music, and more.
This technology operates by learning from large
datasets to generate new, original material that
resembles the learned content. In the context of
text generation, Generative AI uses models like
Large Language Models (LLMs) to produce human-
like text responses based on the input they
receive.
11. Generative AI
• Generative AI is different from other types of AI in text
generation through its ability to create new, original content
based on pattern and examples learned from extensive dataset.
• Content creation vs. Task performance
• Data Driven Learning
• Unsupervised Learning Capabilities
• Generative Models
• Creativity and Adaptability
• Versatility and Content Generation
12. Applications of Generative
AI to Text Generation
• Writing Assistance
• Information Retrieval
• Thought Partnership
• Chatbots and Virtual Assistants
• Language Translation
• Summarisation
• Content Creation for Various Media
13. Class Activity
Generative AI and Chatbots
Divide into groups - each group discuss one area of research (use references/
sources):
• Group 1: Exploring User Preferences: Research how user expectations and
preferences for chatbot interactions are evolving with the use of Generative AI.
• Group 2: Ethical Considerations: Investigate the ethical considerations
surrounding the use of Generative AI in chatbots, such as bias, transparency,
and user privacy.
• Group 3: The Future of Customer Service: Research how Generative AI
chatbots are transforming the landscape of customer service interactions.
• Group 4: Creative Applications: Explore the use of Generative AI chatbots in
creative domains like storytelling, education, or entertainment.
14. Large Language Models
• A language model distinguished by its general-
purpose language generation capability.
• Typically built with a transformer-based
architecture, but some implement recurrent neural
network variants or state space models like Mamba.
• Training Process Acquires abilities through learning
statistical relationships from text documents in a
self-supervised and semi-supervised training
process.
15. How LLMs Work
• Architecture
• Attention Mechanism
• Training Data
• Tokens
• Training Process
16. Capabilities of LLMs
• Text Generation
• Language Translation
• Content Creation
• Question Answering
• Summarisation
• Sentiment Analysis
17. Ethical Challenges and
Considerations for LLMs
• Potential Bias in Training
• Risk of Generating Misinformation
• Privacy Concerns
• Data Security
• Environmental Impact of Training Large Models
• Societal Impact
• Legal and Copyright Issues
• Responsibility and Accountability
• Risks of Malicious Use
• Transparency and Control
18. Online Task
• Choose a specific textual interface (e.g., a virtual
assistant, a search engine) and analyse how it
utilises NLP techniques and/or Generative AI.
• Write a short 300 word description of it, outlining
your observations about it and potential
improvements based on the concepts discussed in
class.