Postgraduate Course: Natural Language Understanding (Level 11) (INFR11061)
|School||School of Informatics
||College||College of Science and Engineering
|Credit level (Normal year taken)||SCQF Level 11 (Year 4 Undergraduate)
||Availability||Available to all students
|Summary||This course explores current research into interpreting natural language. Motivations for this study range from foundational attempts to understand how people interpret communication to entirely practical efforts to engineer systems for performing a variety of language tasks, such as information extraction, question answering, natural language front ends to databases, human-robot interaction and customer relationship management, to name a few.
This course represents an introduction to the theory and practice of computational approaches to natural language understanding. The course will cover common parsing methods for sentences, discourse and dialogue, and it will also address lexical processing tasks such as word sense disambiguation and clustering. We will study state of the art symbolic techniques in deep and shallow language processing, as well as statistical models, acquired by both unsupervised and supervised machine learning from online linguistic resources. Students will have the opportunity to explore what they have learned in written and practical assignments. These assignments will be designed to enable students to gain an understanding for the pervasiveness of language ambiguity at all levels and the problems this poses for automated language understanding, and for the relative strengths and weaknesses of the various theories and engineering approaches to these problems.
* Advanced parsing models; e.g., headed PCFGs
* Grammar Induction
* Discriminative Parsing
* Shallow parsing
* Human models of sentential parsing (e.g., incrementality)
* Semantic Construction in wide-coverage online grammars
* Word sense disambiguation
* clustering, similarity distributions
* lexical subcat acquisition and semantic role labelling
* Human models of lexical processing (e.g., semantic priming)
* Anaphora resolution
* Discourse segmentation
* Dialogue act recognition
* Discourse parsing (including learning discourse structure)
* Human models of discourse and dialogue (e.g., the alignment model)
* Advanced topics
Relevant QAA Computing Curriculum Sections: Artificial Intelligence, Human-Computer Interaction (HCI), Natural Language Computing
Information for Visiting Students
Course Delivery Information
|Academic year 2014/15, Available to all students (SV1)
|Learning and Teaching activities (Further Info)
Lecture Hours 20,
Summative Assessment Hours 2,
Programme Level Learning and Teaching Hours 2,
Directed Learning and Independent Learning Hours
|Assessment (Further Info)
|Additional Information (Assessment)
||Practical exercises, addressing semantic tasks such as word sense disambiguation and discriminative parsing.
You should expect to spend approximately 35 hours on the coursework for this course.
If delivered in semester 1, this course will have an option for semester 1 only visiting undergraduate students, providing assessment prior to the end of the calendar year.
|No Exam Information
| 1 - Given a parsing problem students should be able to use state-of-the-art symbolic parsing techniques, including lexicalised parsing to solve the problem and provide a written explanation of the parsing techniques used in the course.
2 - Given a labelled corpus, students should be able to select and use state-of-the-art statistical parsing techniques (generative and discriminative) by training parsers on the labelled corpus using existing software packages.
3 - Given an NLU system, students should be able to choose appropriate evaluation metrics for the system, and use error analysis to propose improvements to the language processing models.
4 - Given an example of a problem in coreference resolution, discourse segmentation, and discourse parsing, students should be able to provide a written description of how current symbolic and statistical techniques help solve the problem.
5 - Given a description of an NLU system, the student should be able to relate it to features of human models of language interpretation at various levels of processing (words, sentences, discourse and dialogue).
6 - Given a model and a labelled corpus, students should be able to employ existing ML software packages to train the model on the corpus in order to perform a lexical semantic task.
7 - Given an open-ended problem of choosing informative features for a particular NLP task and a description of the available training resources, the student should be able to give a well-justified, written and/or practical, selection of such informative features.
|Speech and Language Processing. An Introduction to Natural Language Processing, Computational Linguistics and Speech Recognition by Daniel Jurafsky and James Martin, Pearson Prentice Hall, 2nd Edition 2008 |
|Course organiser||Dr Frank Keller
Tel: (0131 6)50 4407
|Course secretary||Miss Claire Edminson
Tel: (0131 6)51 4164
© Copyright 2014 The University of Edinburgh - 12 January 2015 4:11 am