Office Hours
- Bhuwan: Wednesdays and Fridays after class (either in Perkins 217 or LSRC D111)
- TAs:
- Monday: 10:00am – 11:00am | Flora Jia (zoom link)
- Tuesday: 10:00am – 11:00am | James Cai (zoom link)
- Wednesday: 6:00pm – 7:00pm | Xunjian Yin (zoom link)
- Thursday: 11:20am – 12:20pm | Xunjian Yin (In-Person, North 232)
- Thursday: 2:30pm – 3:30pm | Flora Jia (In-Person, North 232)
- Friday: 10:00am – 11:00am | Raymond Xiong (zoom link)
Readings
The primary textbooks we will refer to in the course are Eisenstein’s Natural Language Processing and Jurafsky & Martin’s Speech and Language Processing (3rd Ed. draft). For deep learning related content another useful text is Goldberg’s Primer on Neural Networks for NLP. Any other readings relevant for a lecture will be listed under the schedule.
These readings are intended to complement the material discussed in the lecture. You may find it useful to refer to them before or after the class but you are not expected to know things which are not covered in the lectures.
Grading
The final grade will be comprised of:
- 4 assignments (1 x 5% + 3 x 10% = 35%)
- 2 in-class quizzes (2×15% = 30%)
- Weekly exercises (15%)
- Final Project (20%)
The assignments will involve implementing neural network models for common NLP tasks in Python and exploring extensions which improve their accuracy. Quizzes will test you on the mathematical and conceptual understanding of the methods discussed in class. The weekly exercises will consist of 1-3 questions posted on Gradescope each Friday about the material covered that week. They will need to be completed by the following Wednesday. The final project will be conducted in teams of 3-4 students and will involve finetuning an open-source LLM on 2-3 pre-selected benchmarks to maximize its accuracy. We will use the Tinker platform from Thinking Machines Labs for this.
Absences
Students are expected to attend all the lectures without exception. Each week there will be a small exercise posted on Gradescope which will need to be completed by the first class following week. These will be based on the material covered in class that week.
Sickness: To keep the university community as safe and healthy as possible, please do not come to class if you have cold symptoms. Please inform me of your absence and plan to complete any missed work. Students who encounter short- and long-term medical issues or instances of personal distress or emergency can seek academic support if needed. Details regarding university sick leave policies can be found here: Health Issues, Short- and Long-Term.
Policies
Communication: We will use Ed for communication between students and instructors. You can find a link to it at the top.
Late days: We will allow a total of 3 late days on the assignments (cumulatively, not per assignment). You do not need permission to take these. After your allotted late days, you will lose 15% per additional day. If there is an unforeseen health or personal emergency, please contact the instructor.
Collaboration: You may discuss the assignments with other students but all the text and code must be produced independently by you. We will use automated tools for detecting plagiarism. Any violations will be dealt with seriously. As a student, you must abide by the academic honesty standard of Duke University.
Use of AI: This should be treated in the same way as discussion with other students — you can use them to discuss concepts and ideas, and gain a broad understanding or for specific implementation details — but your final code and writing should be your own. You must not use AI tools such as Gemini, Codex or Claude to write program code, inspect your solutions, suggest improvements or contribute to writing the reports. You may use these tools to discuss the concepts behind the assignments and you may use them to proofread your report for grammatical issues. Any such uses, however, must be properly cited and you should note exactly which parts of your submission received AI assistance. Any violations will be strictly punished.
Disabilities: If you are a student with a disability and need accommodations for this class, it is your responsibility to register with the Student Disability Access Office (SDAO) and provide them with documentation of your disability. SDAO will work with you to determine what accommodations are appropriate for your situation. Please note that accommodations are not retroactive and disability accommodations cannot be provided until a Faculty Accommodation Letter has been given to me. Please contact SDAO for more information: sdao@duke.edu or access.duke.edu.
Diversity: We hope that this course will serve students from diverse backgrounds and perspectives equally and respectfully. If you have any concerns or suggestions for improving the effectiveness of this course, please let the instructor know.
