Natural Language Processing

This course introduces students to core and advanced techniques in Natural Language Processing (NLP), with a strong emphasis on practical application across different domains. Students learn to process both written and spoken language using real-world data. Key tasks include emotion classification, speech-to-text transcription, machine translation, semantic representation, and feature extraction. Students gain hands-on experience with both traditional models (e.g., Logistic Regression, Naive Bayes) and deep learning architectures (e.g., LSTM, RNN, Transformers). They apply widely-used NLP libraries and frameworks such as HuggingFace ransformers, SpaCy, NLTK, and Gensim. Feature engineering techniques include POS tagging, TF-IDF, sentiment scoring, and embedding-based representations. The course also covers model evaluation using F1-score, Word Error Rate (WER), and error analysis, as well as explainability methods like Gradient × Input and Layer-wise Relevance Propagation (LRP). Students explore prompt engineering strategies to fine-tune large language models for downstream tasks, and complete the course by designing end-to-end NLP pipelines.

• Text Classification o Logistic Regression, Naive Bayes, LSTM, RNN, Transformers (e.g., BERT, DistilBERT) • Speech-to-Text o Automatic transcription using Whisper and AssemblyAI o Evaluation with Word Error Rate (WER) • Machine Translation o Neural machine translation using pretrained models (e.g., MarianMT) o Round-trip translation and quality assessment • Feature Engineering o Part-of-Speech tagging, TF-IDF, sentiment analysis o Pretrained and custom-trained word embeddings (Word2Vec, GloVe) • Prompt Engineering o Zero-shot and few-shot prompting for classification tasks • Explainable AI for NLP o Gradient × Input o Layer-wise Relevance Propagation (LRP) • Evaluation & Error Analysis o Precision, recall, F1-score, confusion matrices o Qualitative and quantitative assessment of model behavior • End-to-End NLP Pipelines o Modular workflows combining transcription, translation, feature extraction, classification, and explainability