https://studiegids.vu.nl/en/courses/2024-2025/L_PAMATLW007Note: This is a seminar based course with the same learning goals of Advanced NLP. The goal of this course is to gain deep insight in the working of natural language processing methods. In particular, we cover: 1. how to design a system for a complex NLP task; 2. how to analyse the workings of the system through evaluation and interpretability methods;Natural Language Processing (NLP) is a highly dynamic research field that mainly operates on the interface between linguistics and computer science. In order to get computers to deal well with natural language, it is important to understand both how language works and how computational methods work. This course will focus on a complex NLP task (e.g. Semantic Role Labeling, Coreference Resolution) useful in the subfield of Information Extraction. Students will be expected to carry out experiments using supervised models (including traditional, deep learning and transformer architectures). The focus of the course will be on the analysis of model behavior with the goal of identifying both strengths and weaknesses of these models.This course follows a self-study approach. There is a mandatory first lecture to explain how the course will run. Following that, students are expected to revise materials from Advanced NLP independently in order to complete the course requirements. Lecturers will be available for consultation throughout the course.The course has three components:One (possibly oral) midterm exam covering theoretical and practical concept necessary to complete the other two components. This exam will cover: linguistic concepts relevant to the task, theoretical aspects of the machine learning models used in the course and their interaction. This component is worth 20% of the final grade and needs to be a passing grade (at least 5.5).One individual practical assignment covering the implementation of supervised learning systems for the chosen task. This component assumes prior knowledge on how to implement these systems (see ‘entry requirements’ below). This component is worth 40% of the final grade and needs to be a passing grade (at least 5.5).One take home exam focusing on behavioral/black-box testing of the models developed in class. This component assumes prior knowledge in core linguistic concepts in morphology, syntax and semantics. This component is worth 40% of the final grade and needs to be a passing grade (at least 5.5).TBDIn principle the course is targeted at students who have already attempted but did not complete Advanced NLP.In principle the course is targeted at students who have already attempted but did not complete Advanced NLP. The official requirements are the same as the original course: This is an advanced NLP course that assumes a solid background in linguistics, machine learning and NLP. This course requires programming experience and the ability to independently implement and understand supervised machine learning systems (including: feature-based models such as Logistic Regression and SVMs; deep neural methods such as Bi-LSTMs; and transformer-based models such as fine-tuned BERT models). In addition, this course also assumes prior knowledge of core linguistic concepts in morphology, syntax and semantics. Students should be able to understand concepts like syntactic ambiguity, polysemy, and the difference between words and morphemes. Students should be comfortable with constituency or dependency syntactic trees. Students of the Text Mining/HLT programs who have successfully completed all prior courses fulfill all requirements. Other students should ensure they have had a solid course on Machine Learning (covering the range of models mentioned above). In addition, they should also have at least one or two NLP courses where the foundational linguistic concepts mentioned above were covered.Linguistics, machine learning, programming, NLP (basic concepts in these fields will be assumed to be known and not explained). See entry requirements.