1. Course Information

• Instructor: Hanjie Chen

• Semester: Fall 2025

• Time: Thursday 4:00PM - 5:15PM

• Location: KCK 105

• Email: hanjie@rice.edu

• Office Hour: Friday 4:00PM - 5:00PM, DH 2081

2. Course Description

As Natural Language Processing (NLP) continues to advance, the complexity of models has grown exponentially, leading to powerful but often opaque "black-box" systems, like ChatGPT. While these models excel in various tasks, such as commonsense reasoning, machine translation, sentiment analysis, and question answering, their lack of interpretability poses significant challenges in critical applications like healthcare, legal systems, and finance, where understanding the decision-making process is paramount.

The goal of this seminar is to familiarize students with the emerging challenges in NLP and the advancements made in Explainable NLP. This seminar offers an in-depth exploration of the methods and techniques developed to interpret and explain the decision making and working mechanism of NLP models. Key topics include: Feature Attribution Explanation, Hierarchical Explanation, Interpretable and Rationalized Models, Natural Language Explanation, Prompting for LLM Explainability, Human-Centered Explanation, Mechanistic Interpretability, Explanation Evaluation, and Usability of Explainable AI. Students will gain cutting-edge knowledge through lectures and paper readings and hands-on experience through projects.

Throughout this seminar, students will engage with theoretical concepts, practical implementations, and case studies that highlight the importance of interpretability in various NLP applications. The seminar will also address the ethical implications of deploying NLP models, emphasizing the role of interpretability in ensuring transparency and fostering trust in AI systems.

3. Course Format

• Introduction Session. The instructor will give the first two lectures, introducing the course and classic explanation methods.

• Paper Reading. Each week, we will focus on a specific topic in Explainable NLP. Students will be assigned 2 research papers related to the weekly topic, which they are expected to read before class. For each paper, students should formulate 1 discussion question or comment and submit them to Canvas.

• Paper Presentation. In each class, students will give a presentation (20 minutes) on one of the assigned papers. Each student will be responsible for one presentation throughout the semester.

• Discussion Session. After each presentation, there will be a discussion session (15 minutes) focused on the presented paper and the broader topic. Students should at least ask one question or make one comment in each class (counted as part of attendance). The presenter will be responsible for answering these questions and prompting deeper discussion.

• Final Project Presentation (3-credit option). We will hold the final lecture for project presentations. All students, including those enrolled in the 1-credit option, are welcome to attend. Students enrolled in the 3-credit option will give a final presentation on their projects, which should be 7-8 minutes long, followed by 2-3 minutes for Q&A. 

4. Assignments and Evaluation

Please review the Reading List and sign up for paper presentations by the end of September 6th.

• 1-Credit Option

  1. Paper reading and discussion questions: 11 * 4% = 44%

  2. Paper presentation: 40%

  3. Attendance and active participation in paper discussions: 16%

• 3-Credit Option

  1. Paper reading and discussion questions: 11 * 1% = 11%

  2. Paper presentation: 40%

  3. Attendance and active participation in paper discussions: 16%

  4. Final project: 33%

• Rubrics

1. Paper reading and discussion questions. Read the 2 assigned papers each week and submit 1 discussion question (comment) for each paper to Canvas before class.

2. Paper presentation. Each paper presentation should include an introduction/background, research problem/motivation, method, experimental results, and conclusion/takeaway. Presentations will be evaluated based on clarity, understanding of the paper, and the ability to engage the class in discussion.

3. Attendance and active participation in paper discussions. Students are expected to attend the class and ask at least one question or make one comment for each paper presentation.

Grade: 16 (10-11 attendances), 14 (9 attendances), 12 (8 attendances), 10 (7 attendances), 8 (6 attendances), 6 (5 attendances), 4 (4 attendances), 2 (3 attendances), 0 (<3).

4. Final project. The project involves applying interpretability techniques to a chosen NLP task. It will be assessed based on the thoroughness of research, quality of implementation, and effectiveness of the final presentation. Students are encouraged to contact the instructor at the beginning of the course to discuss their project ideas. Additionally, students should schedule regular appointments with the instructor (at least once a month) or attend office hours throughout the semester to provide updates on their progress. 

Final project presentation (33'):

• Introduction (5'): background/motivation, research problem

• Models and datasets (5')

• Methodology (10'): a description of the method

• Experiments (10'): setup, experimental results

• Conclusion (3')

          No final report.

• Grading

The letter grade will be assigned based on the points accumulated:
A+ (98-100), A (94-97), A- (90-93), B+ (87-89), B (84-86), B- (80-83), C+ (77-79), C (74-76), C- (70-73), D+ (67-69), D (64-66), D- (60-63), F (<60)

5. Prerequisites

Students are expected to have completed at least one machine learning course and possess basic knowledge of NLP, including fundamental concepts and common tasks (such as sentiment analysis and question answering). For students opting for the 3-credit hour option, which includes hands-on projects, programming experience, preferably in Python, and familiarity with libraries such as NumPy, pandas, and scikit-learn, are required.

6. Honor Code

All students are expected to adhere to the standards of the Rice Honor Code, which you agreed to uphold upon matriculation. For detailed information on the Honor Code, including its administration and the procedures for addressing alleged violations, please refer to the Honor System Handbook available at http://honor.rice.edu/honor-system-handbook/. This handbook outlines the University’s expectations for academic integrity, the procedures for resolving any alleged violations, and the rights and responsibilities of students and faculty throughout the process.

Note that students should use AI tools carefully in their assignments. Please follow the university's AI Usage Guidelines.