Statistical Structure in Language Processing

Time: Tuesday 12:00- 14:00, Place: REC-D 3.17

Objectives

The course aims at providing the student with enough skills and background for conducting research in the field of statistical Computational Linguistics/ Natural Language Processing.

Course Description

The amount of language data that is available to us electronically is increasing with the day. With this eminent increase, a question arises as to the possibility of inducing latent structure in this data that can be useful for further tasks such as machine translation. The different kinds of latent structure that is possible depends on the data and the task, and will usually demand suitable statistical models and learners. The course will study methods for inducing a variety of latent structure for tasks such as language modeling, machine translation and adaptation across domains.

The course is divided into two main parts:

• In the first part, we will deal with induction of statistical structure from a mono-lingual point of view. This part will cover different aspects of inducing linguistic structure from incomplete data, like estimation from completely unlabeled data, estimation from partially-labeled data, and learning automatic treebank-refinements (i.e. different levels of annotation). We will also look at estimation for different linguistic models such as dependency grammars and Data-Oriented Parsing (estimation over entire subtrees).

• The second half of the course will focus entirely on Statistical Machine Translation, covering classical and current influential SMT models, evaluation schemes for translation, Synchronous grammars, and, if time permits, maximum-entropy models in SMT.

The course takes the form of an advanced seminar: there will be some lectures by guest lecturers, but a large portion of the course will focus on reading and discussing current literature related to the various topics described above. Students will present papers and lead discussions. Students are also expected to complete a substantial project on one of the topics of the course, depending on their interest.