This course builds upon the foundational principles of computational linguistics to explore the state-of-the-art deep learning models that power modern natural language processing (NLP). The focus is on understanding, interpreting, and critically evaluating these models from a linguistic perspective. The course delves into vector-space models of meaning (word embeddings such as word2vec and GloVe), the architecture of neural networks like Transformers (e.g., BERT, GPT), and their application to complex tasks in linguistics and machine translation. Special emphasis is placed on probing the models' knowledge about language, their inherent biases, and their practical use in areas such as sentiment analysis, machine translation (and its quality estimation), cross-lingual analysis, and other common text generation tasks. Through hands-on projects, students will learn to fine-tune pre-trained models and apply them to novel NLP tasks and research problems.

Note on the Use of Generative AI (GenAI): The use of GenAI tools is encouraged as a productivity aid for brainstorming, debugging code, and improving writing. However, students are expected to critically evaluate all GenAI output. Any text or code generated by these tools and incorporated into assignments must be explicitly acknowledged and cited. The final research project, in particular, should demonstrate the student's original thought and critical analysis, which goes beyond the capabilities of current GenAI. Submitting unedited, unattributed GenAI output as one's own work constitutes academic dishonesty.