TL;DR
This paper outlines Georgetown University's graduate-level Computational Linguistics (CL) and Natural Language Processing (NLP) curriculum, emphasizing a balanced approach that caters to diverse student backgrounds and goals. It highlights the importance of practical data-driven skills, interdisciplinary collaboration, and the inclusion of languages beyond English, offering a model for other institutions developing similar programs.
What Problem Does This Paper Address?
The authors address the challenge of designing and implementing a comprehensive and effective graduate curriculum in Computational Linguistics (CL) and Natural Language Processing (NLP). This is particularly relevant given the rapid growth of these fields and the diverse academic backgrounds and career aspirations of students entering CL/NLP programs. Specifically, they aim to create a curriculum that:
- Accommodates students with varying levels of programming experience.
- Provides a strong foundation in both theoretical concepts and practical, data-driven skills.
- Fosters interdisciplinary collaboration between linguistics and computer science students.
- Encourages the study of languages beyond English, acknowledging the limitations of English-centric NLP.
- Prepares students for a range of career paths in academia and industry.
Methods
The paper doesn't describe novel algorithms or techniques in the traditional sense. Instead, it details the pedagogical methods and curriculum design principles employed at Georgetown University. Key aspects include:
- Sequenced Course Offerings: A multi-course NLP sequence designed to gradually introduce programming and NLP concepts, allowing students with no prior programming experience to start with an "Introduction to NLP" (INLP) course, while those with strong programming backgrounds can begin with "Empirical Methods in NLP" (ENLP). An "Advanced Python for CL" course bridges this gap.
- Interdisciplinary Cross-listing: Many courses are cross-listed between Linguistics and Computer Science departments to encourage collaboration and broaden student access.
- Focus on Data: Emphasis is placed on working directly with language data, including corpus design, annotation, and analysis.
- Project-Based Learning: Courses often involve hands-on projects, including research projects that produce linguistically annotated data, such as contributions to the Georgetown University Multilayer (GUM) Corpus.
- Mock Review Activities: To develop research skills, some courses include mock peer-review simulations of ACL-style submissions.
- Special Topics Courses: These courses delve into emerging or specialized areas, often reading- and research-oriented, covering topics like social factors and ethics in NLP, discourse parsing, and grammar formalisms.
Datasets
The paper does not focus on specific datasets for training or evaluation in the typical machine learning sense. However, it highlights the creation and use of:
- Georgetown University Multilayer (GUM) Corpus: A resource developed through student projects, providing multilayered annotations for various genres. This corpus is intended to be released to the research community under an open license.
Key Insights and Ideas
- Catering to Diverse Backgrounds is Crucial: Acknowledging that students arrive with vastly different levels of programming proficiency and academic focus (linguistics vs. computer science) is paramount for curriculum design.
- Bridging the Programming Gap: The introduction of an "Advanced Python for CL" course is a key innovation to ensure students with less programming experience can successfully engage with more advanced NLP topics.
- Data-Centricity is Non-Negotiable: Emphasizing hands-on work with real language data, including corpus creation and annotation, is essential for developing practical NLP skills.
- Interdisciplinarity Enhances Learning: Cross-listing courses and encouraging mixed-background project groups fosters respect, broader perspectives, and richer learning experiences.
- Beyond English is Essential: Actively incorporating data and examples from languages other than English is vital to address the inherent biases in English-centric NLP and prepare students for a globalized field.
- Social Impact and Ethics are Growing Concerns: The recent introduction of a "Social Factors in NLP" course reflects a growing awareness of the need to address societal impacts, bias, and ethical considerations within the curriculum.
Key Results
The paper does not present quantitative results in terms of model performance. Instead, its "results" are qualitative and pedagogical:
- Successful Curriculum Development: The paper demonstrates the successful implementation of a graduate CL/NLP curriculum at Georgetown University that has seen significant growth over seven years.
- Student Engagement: The curriculum is designed to engage students with diverse backgrounds, from those with no programming experience to seasoned computer scientists.
- Production of Annotated Data: Student projects contribute to valuable linguistic resources like the GUM Corpus.
- Preparation for Diverse Careers: The curriculum aims to equip students for various roles in academia and industry.
Limitations and Future Work
- Limitations:
- The curriculum primarily caters to graduate students, with undergraduate offerings being occasional.
- Teaching programming skills alongside NLP fundamentals remains a significant challenge, requiring substantial support from teaching assistants.
- While progress has been made, integrating societal impact and ethical considerations into core NLP courses is an ongoing effort.
- Future Work:
- Develop a more consistent and structured NLP curriculum for undergraduates.
- Further integrate content on societal impact and ethical considerations into core NLP courses.
- Expand the curriculum to cover other underrepresented topics such as grammar engineering and computational psycholinguistics.
- Re-offer dedicated courses on parallel and multilingual corpora, contingent on resource availability.
Summary updated