CSCI 544 · Applied NLP

Contact

Students should ask all course-related questions in the Ed forum, where you’ll also find announcements. The course Ed page is linked from the Brightspace page. TA office-hour Zoom links are available on Brightspace/Ed. For external enquiries, emergencies, or personal matters you don’t wish to put in a private Ed post, email csci544-25f@googlegroups.com with CSCI544 in your subject line. Please send all emails to this mailing list — do not email the instructors directly. We try to respond within 48–72 hours.

Assignments & Grading

• Homeworks (20%) — 10% × 2: Two coding homework assignments based on class topics.
• Quizzes (10%) — 5% × 2: Multiple-choice and short-answer questions. Missed quizzes receive a zero; no make-up quizzes.
• Class Projects (50%) — Each student does a group project based on class topics. Students propose their own project, do the research and build a proof-of-concept, create a video demonstration, and present the project in their report.
  • Proposal: 5%
  • Status Reports: 10%
  • Project Presentation: 15%
  • Final Write-up: 20%
• Exams (20%) — Midterm (20%): Mixture of multiple-choice and long-form questions, covering the first half of the material.

Grading inquiries can be asked (to the TAs) within two weeks from the grading date. Grades will be available within 2–2.5 weeks after submission.

All written assignments related to the final project should use the standard *ACL paper submission template.

Project Deliverables

Project proposal (5%)

Student teams should submit a ~1-page proposal (using the *CL paper submission template) for their project. The proposal should:

• state and motivate the problem by providing a problem or task definition (preferably with example inputs and expected outputs),
• situate the problem within related work (this might help you find sources of data for training a model for your task),
• include related work / publications — start by looking in the ACL anthology,
• references do not count towards page limit, but please follow the correct format,
• state a hypothesis to be verified and how to verify it (evaluation framework), and
• provide a brief description of the approach to be followed to verify the hypothesis (such as proposed models and baselines).

We highly encourage students to work towards a problem involving predictive models. It’s worth thinking about the five key ingredients of supervised learning: data, model, loss function, optimization algorithm, and inference / evaluation.

Project progress report (10%)

Student teams should submit a ~3-page progress report (using the *CL paper submission template). This report should:

• once again describe the project’s goals (it’s okay if this has changed slightly since the proposal, based on feedback),
• contain all details on the dataset (your dataset should mostly be collected by this time),
• contain some initial results (motivating results), and
• outline a concrete plan of what will be done before the final report.

While the initial results might be inconclusive, you are expected to have made non-trivial progress by this point. The project proposal may be extended for this report. Please take into consideration the earlier feedback you received, and address those inline.

Project final presentation (15%)

Each team will prepare a 5-minute presentation, followed by 1-2 minutes of Q&A. You can choose a representative for your team. Use one slide to clearly describe which team member is responsible for which part. Points will be deducted if the time limit is violated, so please practice timing your talk.

Each presentation should describe:

• the underlying motivation of the project,
• the research questions answered in the project,
• the proposed methods,
• the findings so far,
• the contribution of each member, and
• audience questions answered.

If you are in the audience, you can earn bonus points by asking insightful questions (and clearly announcing your name before asking).

Each team will prepare slides (via Google Slides) and add the link in the course presentation spreadsheet shared through Brightspace/Ed by 11:59 PM the day before the presentation. Failure to share slides on time will cause a loss of grade.

Project final report (20%)

Student teams should submit a ~4-6 page final report (using the *CL paper submission template) detailing all aspects of their project. The report should be structured like a conference paper, including:

• an abstract,
• an introduction to the problem and method,
• related work, highlighting similarities and differences to your own work,
• a description of the method used,
• the experiments and results, and
• a discussion of the results, outlining future work possibilities.

A tech report format is discouraged. Parts of the proposal and progress report may be reused for the final report. Negative results will not be penalized, but should be accompanied with detailed analysis of why the proposed method did not work as anticipated. You may include an appendix at the end. References and the appendix do not count towards the main report page limit. You must submit all your code as a final deliverable (zip file). Plagiarism will be strictly penalized.

Late Days

Students are allowed a maximum of 4 late days total for all assignments (but NOT the quizzes). You may use up to 2 late days per assignment. Using one late day for a project assignment involves each teammate using a late day. Partial late days are not permitted. For every extra late day beyond the allowed limit, the student/team will lose 20% of the grade for that assignment.

Please familiarize yourself with the academic policies and read the note about student well-being.

Reference Texts

These texts are useful but none are required. All are free to read online.

• Dan Jurafsky and James H. Martin — Speech and Language Processing (2024 pre-release)
• Jacob Eisenstein — Natural Language Processing
• Yoav Goldberg — A Primer on Neural Network Models for NLP
• Ian Goodfellow, Yoshua Bengio, and Aaron Courville — Deep Learning
• Delip Rao and Brian McMahan — Natural Language Processing with PyTorch (requires Stanford login)
• Lewis Tunstall, Leandro von Werra, and Thomas Wolf — Natural Language Processing with Transformers
• Background refreshers:
  • Michael A. Nielsen — Neural Networks and Deep Learning
  • Eugene Charniak — Introduction to Deep Learning