Assignments

Overview

In this assignment, you will build a language identification classifier that distinguishes between six languages:

• Hausa
• Indonesian
• Manobo
• Nahuatl
• Swahili
• Tagalog

These six languages use the same (Latin) script with minimal diacritics, making it challenging to distinguish them without sophisticated NLP techniques.

Learning Objectives

• Implement Multinomial Logistic Regression from scratch
• Work with character-level n-gram features
• Evaluate classification performance using confusion matrices
• Analyze feature importance in text classification

Tasks

1. Implement a training loop for Multinomial Logistic Regression
2. Implement inference for Multinomial Logistic Regression
3. Determine the optimal order of n for character n-grams
4. Generate and analyze a confusion matrix for model evaluation
5. Identify and display the most predictive features for each language

Overview

In this assignment, you will build your first language models, exploring both classical and neural approaches to language modeling.

Learning Objectives

• Implement n-gram language models with smoothing
• Build recurrent neural network language models
• Compare different approaches to language modeling
• Handle out-of-vocabulary words effectively

Tasks

1. Implement n-gram language models with Laplace smoothing
2. Build Transformer-based language models
3. Evaluate perplexity across different model architectures
4. Analyze model performance on various text domains

Status

In this homework, you will be experimenting with Neural Machine Translation (NMT) using sequence-to-sequence transformer models. You will fine-tune pre-trained MarianMT models to translate English sentences into French, compare model architectures, and tune hyperparameters to improve translation quality. You will also perform an in-depth error analysis on your model's outputs using BLEU score as your evaluation metric.

In this assignment, you will be fine-tuning pre-trained translation models using two different MarianMT configurations.

Status

Assignment details will be announced later in the semester.

Exam Format

Both exams will consist of 3-4 multipart questions designed to test conceptual understanding rather than memorization. The exams are open book and open notes but non-collaborative.

Topics Covered

• Midterm: Language modeling, classification, neural networks, word embeddings
• Final: Comprehensive, with emphasis on sequence modeling, attention, transformers, and applications

Preparation

Review lecture materials, complete practice problems, and ensure you understand the conceptual foundations behind each technique covered in class.