Course development report 2024
This is the 2024 course development report for TDDE09 Natural Language Processing (6 credits).
Statistics
The 2023 session had 39 registered participants. Based on social security numbers, 32 out of these were men and 7 were women. There were 3 exchange students (France, The Netherlands). The examiner was me, Marco Kuhlmann.
After the first examination (2023-03-25), 27 out of 39 students (69%) had passed the course. This number had increased to 32 out of 39 students (82%) after the last examination (2023-08-27).
Course evaluation
The course evaluation for the 2023 session was open between 2023-03-13 and 2023-04-09. It received responses from 15 out of 41 respondents (response rate: 37%). The overall evaluation score of the course was 4.80 out of 5 (median: 5.00).
Below is a summary of the free-text replies.1
What students like about the course
Quality of Course Material and Teaching: Students are highly appreciative of the course material and the teaching approach. They find the labs well-structured, challenging, and fun, contributing significantly to their learning experience. Students also highlight the relevance and richness of the course material, with special mention of its up-to-date nature in relation to current research.
Engaging Learning Methods: Students commend the use of quizzes, the project, and the reflection paper. Quizzes are seen as a good motivational tool to encourage attentive listening during lectures. The project is considered interesting, enhancing the overall learning experience. Students appreciate the reflection paper for facilitating deeper understanding of the project.
Positive Course Atmosphere and Teaching Style: Students express satisfaction with the overall atmosphere of the course, facilitated by the examiner’s approachable style and additional touches like snacks during project presentations. This positive environment, along with the pedagogical effectiveness of the instructor, contributes to a favorable learning experience.
What students think can be improved
Course Design and Content: Students appreciate the overall course design and find the content interesting. They commend the use of quizzes for reinforcing learning but suggest that quizzes should focus more on key concepts rather than tricky details. They also suggest incorporating more current topics like GPT in labs and offering more interactive elements to enhance understanding.
Workload and Time Management: Many students find the course demanding in terms of time and effort. They mention that the workload seems excessive for the credit value of the course, and balancing the course with other commitments can be challenging. Some suggest reducing the number of quizzes and the length of labs to make the course more manageable.
Learning Methods and Assessment: There is a desire for a better balance between different learning methods, such as videos, lectures, and reading materials. Students also raise concerns about the difficulty in adapting to new tools like PyTorch and the stress associated with writing academic essays, which is unfamiliar to many engineering students.
Examiner comments
Natural language processing is a fast-moving field, and it is not easy to cover both the basic concepts and up-to-date topics in one 6-credits course, especially given the heterogeneous student population2. I am therefore happy to see that students appreciate the course content, design and atmosphere.
I am grateful to the students for their constructive feedback. I have tried to integrate this into the course development ahead of the 2024 session as follows.
Course Design and Content: I have restructured the course content to put more emphasis on current topics. To this end, I have compressed the previous lecture units sequence labelling and syntactic parsing into a single unit structured prediction, which has freed up time that I plan to spend on current research. I have also changed the order of units so that language modelling now comes before word representations. This switch reflects the current interest in language models and also means that the challenging lab on word representations comes later in the course, which should make for a more gentle learning path when it comes to PyTorch.
Workload and Time Management: Student assessments of workload has varied over the last few years. For the 2022 session, 81% of respondents considered the workload adequate and only 6% considered it too high. For the 2023 session, these numbers have changed to 53% versus 47%. The main change between 2022 and 2023 was the introduction of the quizzes. I hope that the changes to the course structure (explained above) and the quizzes (explained below) will revert some of the increase in perceived workload.
Learning Methods and Assessment: The main purpose of the quizzes is to motivate students to actively engage with the content of the video lectures. However, some of the quizzes required more than just recalling facts from the videos, which some students considered to be a too-high level of understanding for this type of assessment. I have moved most of these more challenging elements to a new category of in-class assignments, which are administered during the on-campus sessions. I hope that this makes for a better alignment between intended level of knowledge and assessment format.