MDS Computational Linguistics
UBC’s new Master of Data Science program with Computational Linguistics Specialization is the credential to set you apart. Offered at the Vancouver campus, this unique degree is tailored to those with a passion for language and data. Over 10 months, the program combines foundational data science courses with advanced computational linguistics courses—equipping graduates with the skills to turn language-related data into knowledge and to build AI that can interpret human language.
Highlights Across All MDS Programs:
- 10-month, full-time, accelerated program offers a short-term commitment for long-term gain
- Condensed one-credit courses allow for in-depth focus on a limited set of topics at one time
- Capstone project gives students an opportunity to apply their skills
- Real-world data sets are integrated in all courses to provide practical experience across a range of domains
Highlights Specific To Computational Linguistics:
- Courses are taught by a combination of arts (linguistics), computer science, and statistics faculty members giving students access to key experts within each field of study
- Students learn fundamental data science skills, techniques, and tools with the core Master of Data Science cohort, then branch off into more specialized courses, experiencing the benefits of a large program and small program in one
- UBC’s Vancouver campus offers students the unrivaled experience of a top 40 university, surrounded by remarkable natural beauty, at the edge of a cosmopolitan city
- Strong connections with industry partners in public and private sectors, start-ups, and leading tech companies offer a wide range of networking/career opportunities
The program structure includes 24 one-credit courses offered in four-week segments. Courses are lab-oriented and delivered in-person with some blended online content.
At the end of the six segments, an eight-week capstone project is also included, allowing students to apply their newly acquired knowledge, while working alongside other students with real-life data sets. Please note that instructors are subject to change.
Fall: September - December
Block 1 (4 weeks)
How to install, maintain, and use the data scientific software “stack”. The Unix operating system, integrated development environments, and problem solving strategies.
Fundamental concepts in probability. Statistical view of data coming from a probability distribution.
Basic processing of text corpora using Python. Includes string manipulation, corpus readers, linguistic comparison of corpora, structured text formats, and text preprocessing tools.
Block 2 (4 weeks)
Converting data from the form in which it is collected to the form needed for analysis. How to clean, filter, arrange, aggregate, and transform diverse data types, e.g. strings, numbers, and date-times.
Exploratory data analysis. Design of effective static visualizations. Plotting tools in R and Python.
How to choose and use appropriate algorithms and data structures to help solve data science problems. Key concepts such as recursion and algorithmic complexity (e.g., efficiency, scalability).
The statistical and probabilistic foundations of inference, developed jointly through mathematical derivations and simulation techniques. Important distributions and large sample results. Methods for dealing with the multiple testing problem. The frequentist paradigm.
Block 3 (4 weeks)
Linear models for a quantitative response variable, with multiple categorical and/or quantitative predictors. Matrix formulation of linear regression. Model assessment and prediction.
The identification of syntactic structure in natural language. Parsing algorithms for popular grammar formalisms, application of statistical information to parsing, parser evaluation, and extraction of parse features.
Introduction to supervised machine learning, with a focus on classification. K-NN, Decision trees, SVM, how to combine models via ensembling: boosting, bagging, random forests. Basic machine learning concepts such as generalization error and overfitting.
How to work with data stored in relational database systems. Storage structures and schemas, data relationships, and ways to query and aggregate such data.
Winter: January - April
Block 4 (4 weeks)
How meaning is represented by computers. An overview of popular semantic resources, and techniques for building new resources from unstructured text data.
How to evaluate and select features and models. Cross-validation, ROC curves, feature engineering, and regularization.
How to find groups and other structure in unlabeled, possibly high dimensional data. Dimension reduction for visualization and data analysis. Clustering, association rules, model fitting via the EM algorithm.
Introduction to optimization. Gradient descent and stochastic gradient descent. Roundoff error and finite differences. Neural networks and deep learning.
Block 5 (4 weeks + 1 week break)
The legal, ethical, and security issues concerning data, including aggregated data. Proactive compliance with rules and, in their absence, principles for the responsible management of sensitive data. Case studies.
Approaches to sub-word phenomenon in language processing. Automatic morphological analysis of diverse languages, part of speech tagging, word segmentation, and character-level neural network models.
Key methodologies for automatic translation between languages, with a focus on statistical and neural machine translation approaches. Applying Machine Translation (MT) architectures to analogous monolingual tasks. MT evaluation.
Text corpora collection and curation. How to pull representative datasets from internet sources. Techniques for efficient and reliable annotation.
Block 6 (4 weeks)
Application of machine learning to various semantic tasks. Likely topics include: information extraction, semantic role labelling, semantic parsing, discourse parsing, question answering, summarization, and natural language inference.
Cutting-edge techniques in natural language processing. For this iteration, the latest innovations in neural network architectures.
Identification and analysis of opinion, especially in social media. Text polarity and emotion classification, fine-grained (e.g. aspectual) opinion mining, argumentation mining, sentiment in social networks.
Building automatic language tools when data is scarce. Rule-based and hybrid systems, semi-supervised learning, active learning. Knowledge transfer from other (related) languages.
Spring: May - June
Capstone Project (8-10 Weeks)
A mentored group project based on real data and questions from a partner within or outside the university. Students will formulate questions and design and execute a suitable analysis plan. The group will work collaboratively to produce a project report, presentation, and possibly other products, such as a web application.
Data in Action: Helping AI Recognize Intent
As part of their capstone project, students from UBC’s Master of Data Science program partnered with Finn Ai, to help the banking software company improve their AI assistant’s ability to identify user intents.
Examining the company’s existing neural network model, the students were able to identify areas of confusion for the AI and improve customer service.