Computational Linguistics
Lecturer:
Prof. Daniel R. Schlegel, 464 Shineman Center, daniel.schlegel@oswego.edu
Office/Lab hours: Typically 2-3pm on Zoom Monday through Friday, but send mail to set up an appointment or ask questions any time.
Section 800: TTh 11:20am-12:40pm, Shineman 175
Course Description:
This course provides an introduction to natural language processing techniques. Specification, implementation, and evaluation of machine learning techniques as applied to natural language will be discussed. We will examine relevant linguistic constructs as we build from the bag of words model of language to richer structural models representing the relationships between words and phrases to encode meaning.
Course Objectives:
Students who complete this course will be able to:
- Grasp fundamental concepts in linguistics relevant to natural language processing.
- Understand and discuss the mathematical and theoretical underpinnings of algorithmic and statistical techniques applied to language processing.
- Select approaches for solving language processing tasks and defend their decisions.
- Implement algorithms and techniques relevant to language processing applied to problems addressed by the field.
- Evaluate performance of algorithms on linguistic tasks and improve upon results using knowledge of the algorithms, linguistics, and the problem being solved.
Prerequisites:
The course catalog prerequisite for this course is CSC241. I would add that a course in calculus and some programming experience beyond CSC241 is very highly recommended. Ideally, you will have taken CSC365. This course will make use of calculus, linear algebra, probability and statistics. Most everyone will have to pick up some additional mathematical skills along the way, but having to pick up all of them is probably not possible.
Textbooks:
Required: Eisenstein, J. Introduction to Natural Language Processing. MIT Press, 2019. [See Blackboard!]
Useful Resources:
Jurafsky and Martin – Speech and Language Processing 3rd Edition Draft
The Python Tutorial
PyTorch Tutorials
NumPy Documentation
Attendance and Participation:
As per college policy, attendance in all sessions is obligatory. If you cannot attend a class meeting due to religious, athletic, health related circumstance, or circumstance of particular hardship, please notify me in advance via email. Please be ready to present proof, if necessary. It is expected that each person actively engage in each class session.
This course includes a significant discussion component. Participation in discussions is mandatory and will be factored into the final grade.
Classroom Etiquette:
A positive learning environment relies upon creating an atmosphere where all students feel welcome. Classroom discussion is meant to allow us to hear a variety of viewpoints. This can only happen if we respect each other and our differences. Hostility and disrespectful behavior is not acceptable.
Proper mask etiquette and social distancing must be observed in the classroom at all times!
Grading Summary:
Grades will be comprised of participation, programming projects, written homework assignments, biweekly quizzes, and a final exam. A point-based system will be used, where each graded artifact will be assigned a point value and you can simply sum the points to determine your grade.
| Assessment | Points |
|---|---|
| Projects (3-4) | 400 |
| Homeworks (7) | 70 |
| Participation | 130 |
| Quizzes (5) | 200 |
| Final Exam | 200 |
| Total | 1000 |
The default grading for the course will be along the university’s standard grading curve:
| Letter: Points | Letter: Points |
|---|---|
| A: 930-1000 | C+: 770-790 |
| A-: 900-920 | C: 730-760 |
| B+: 870-890 | C-: 700-720 |
| B: 830-860 | D+: 670-690 |
| B-: 800-820 | D: 600-660 |
| E: 0-590 |
Projects:
All projects are to be completed alone and submitted on Blackboard once complete. Be sure not to post solutions on the internet during or after the course as we wish to use these problems in the future.
Projects will be graded based on completion and quality of submission (including quality of code). All projects have a competitive component in which points will be assigned for particularly good solutions scored objectively on hidden data sets. Results will be presented in class and particularly interesting solutions may be examined in detail.
Projects are considered on-time if they are submitted on or before the due date, with an 11:59pm cutoff time for submission. Projects may still be submitted after the deadline with a 5% per day penalty.
Note that no credit will be given for projects which fail to run, and partial credit will be given if only parts of the project work as described.
Homework Assignments:
Homework assignments will give you additional practice with some of the more theoretical concepts discussed in class. Solutions are to be written on the provided homework sheets and submitted on the due date at the start of class. No late homework assignments will be accepted.
Homework due dates correspond to the quiz which will test (among other things) the understanding of concepts from the homework assignments.
Exams and Quizzes:
You may bring your book to quizzes and the final exam, but may not use any notes or electronic aides.
Quizzes will be given roughly every two weeks, and there will be a take-home final exam during finals week. The lowest quiz grade will be dropped.
Each exam and quiz question will be assigned a point value, questionPoints, where the following general scheme will be used in grading it:
0 – Did not attempt / No serious attempt / Completely incorrect
1/3 * questionPoints – Mostly incorrect solution
2/3 * questionPoints – Somewhat incorrect solution
3/3 * questionPoints – Perfect solution
Intermediate scores will be given as appropriate. The total points received on all questions will then be summed.
Schedule/Outline:
The course will be generally divided into four segments, during which we will build up our understanding of algorithms applied to the structure of language from simple word frequency based models to semantics based on structure. This outline is detailed in the graphic syllabus. This is highly optimistic, and we may not get through everything.
This syllabus and the course schedule are subject to change by the instructor. All changes and related justifications will be announced in class, and updates will be reflected in this web version.
| Week | Day | Date | Topic | Reading Due | Assignments/Assessments |
|---|---|---|---|---|---|
| 1 | Tuesday | 2/2 | First day of class Course Overview | Pretest | |
| Thursday | 2/4 | Finish Overview Discuss Pre-test | Chapter 1, Appendix A See (Goldwater, 2018) for more probability help. | HW1 Assigned | |
| 2 | Tuesday | 2/9 | No Class | ||
| Thursday | 2/11 | Bag of Words Preprocessing | Chapter 2-2.1, Chapter 4-4.1 (Discussion Q's on Blackboard) | HW1 Due Project 1 Assigned |
|
| 3 | Tuesday | 2/16 | No Class | 2.2 (Naive Bayes), 4.2 (Word Sense Disambiguation) (Discussion Q's on Blackboard) | |
| Thursday | 2/18 | Naive Bayes | HW2 Assigned Quiz 1 Assigned (Take Home - Due Tuesday) |
||
| Friday | 2/19 | Drop deadline | |||
| 4 | Tuesday | 2/23 | Naive Bayes Example Perceptron | 2.3-2.4 (Discriminative Learning), 4.3 (Design Decisions for Classification) (Discussion Q's on Blackboard) | |
| Thursday | 2/25 | Perceptron | Appendix B (Optimization) 2.5-2.6 (Logistic Regression / Optimization) (Discussion Q's on Blackboard) | HW2 Due Quiz 2 Due Tuesday |
|
| 5 | Tuesday | 3/2 | Perceptron Logistic Regression | ||
| Thursday | 3/4 | Logistic Regression Gradient Descent | |||
| 6 | Tuesday | 3/9 | Gradient Descent Example Neural Nets Intro | 3-3.3 - Neural Networks You may find 7.3-7.4 in the slp3 draft helpful | |
| Thursday | 3/11 | Neural Nets, Backprop | HW3 Due Quiz 3 |
||
| 7 | Tuesday | 3/16 | Part of Speech + Sequence Labeling UD POS Labels | 7-7.1, 8 - Sequence Labeling + Applications | Project 1 Due |
| Thursday | 3/18 | Hidden Markov Models Viterbi Algorithm | 7.2-7.4 - HMMs | Project 2 Assigned | |
| 8 | Tuesday | 3/23 | No Class - Wellness Day | ||
| Thursday | 3/25 | n-gram Language Models | 6-6.5 - Language Models | ||
| 9 | Tuesday | 3/30 | Neural Language Models RNNs and LSTMs Illustrated guide to LSTMs and GRUs | 7.5-7.6 - Discriminative Models, Neural Sequence Models | |
| Thursday | 4/1 | Word Embeddings | 14-14.4 - Distributed Semantics | HW4 Due Quiz 4 |
|
| 10 | Tuesday | 4/6 | Word Embeddings + Semantics Explore Embeddings! More Embeddings to Explore! Phrase-Structure Grammar and Constituency Parses Progress Report / Group Work | 14.5 - 14.8 - Neural Embeddings Chapter 9 - Formal Language Theory | |
| Thursday | 4/8 | CKY Parsing | 10-10.3 - Bottom-Up Parsing | ||
| Monday | 4/12 | Withdraw Deadline | |||
| 11 | Tuesday | 4/13 | Work Day! | ||
| Thursday | 4/15 | Shift-Reduce Parsing / Dependency Parsing | 11-11.3 - Dependency Parsing | HW5 Due Quiz 5 |
|
| 12 | Tuesday | 4/20 | Transition-Based Dependency Parsing Project 3 Introduced First Order Logic Semantics | 12-12.2 - Logical Semantics | Project 2 Due Project 3 Assigned (Due last day of finals week) |
| Thursday | 4/22 | No Class | |||
| 13 | Tuesday | 4/27 | First Order Logic Semantics | ||
| Thursday | 4/29 | Semantic Parsing | 12.3-12.4 - Lambda Calculus | ||
| 14 | Tuesday | 5/4 | Predicate-Argument Semantics | 13 - Predicate-Argument Semantics | |
| Thursday | 5/6 | A Real-Life Actually Implemented (Domain-Limited) Language Understanding System Course Evals | Natural Language Understanding for Soft Information Fusion | HW6 Due | |
| Finals Week | Take-Home Exam Assigned Monday, due Friday 4:30pm | Final Exam |
Academic Integrity:
SUNY Oswego is committed to Intellectual Integrity. Any form of intellectual dishonesty is a serious concern and therefore prohibited. You can find the full policy online. While it is acceptable to discuss general approaches with your fellow students, the work you turn in must be your own. You may not turn in code found on the internet. If you have any problems doing the assignments, consult the instructor. See my page on plagiarism for an explanation of what I consider cheating.
Accessibility:
If you have a disabling condition which may interfere with your ability to successfully complete this course, please contact Accessibility Resources located at 155 Marano Campus Center, phone 315.312.3358, access@oswego.edu
Clery Act/Title IX Reporting:
SUNY Oswego is committed to enhancing the safety and security of the campus for all its members. In support of this, faculty may be required to report their knowledge of certain crimes or harassment. Reportable incidents include harassment on the basis of sex or gender prohibited by Title IX and crimes covered by the Clery Act. For more information about Title IX protections, go to https://www.oswego.edu/title-