DTAI

Tuesday December 20 2011, at 16h30 in Celestijnenlaan 200A (room 05.001)

Kernel-based Logical and Relational Learning of Natural Language
by Mathias Verbeke (PhD student DTAI)

While there has been a lot of progress over the past decade and promising results have been obtained on a wide variety of language learning tasks such as
part-of-speech tagging, parsing, etc., other tasks remain extremely challenging because solving these tasks requires one to combine syntactic with semantic dependencies, structured with unstructured data and local with global models. It is unclear how to realize this using state-of-the-art language learning techniques. On the other hand, statistical relational learning (SRL), with its integrated approach to logic and learning, may well be able to solving these problems.

In this seminar, the first step in this deeper understanding on how to apply logical learning techniques for computational linguistics will be explained.
kLog, a new logical and relational language for kernel-based learning will be introduced. Following this, we will illustrate our approach for hedge cue
detection, a natural language processing task that consists of determining whether sentences contain hedges, i.e. linguistic devices that indicate that
authors do not or cannot back up their opinions or statements with facts. In particular, we will focus on the advantage of the SRL approach in general, and kLog in particular, of being able taking into account additional (linguistic) background knowledge and context.