• Home
  • Text Technology Lab: Three guest talks

Text Technology Lab: Three guest talks

EVENT


CLT Text Technology Laboratory seminar with three guest talks

13.30-14.10 Gerlof Bouma (University of Potsdam) 

14.15-14.55 Simon Dobnik (Oxford University) 

14.55-15.10 break

15.10-15.50 Richard Johansson (University of Trento)

Abstracts:

Gerlof Bouma (University of Potsdam): Association measures assorted

In this talk I will discuss extensions of pointwise mutual information and (average) mutual information, two classic association measures popular in the field of collocation extraction. The essence of both measures is the comparison of observed occurrence of a combination and its expected occurrence under the assumption that the parts in the combination are independently distributed. The first extension tries to reduce the effect of occurrence probability and to improve the interpretability of the association measures by normalization. In the second extension, we relax the independence assumption in the calculation of expected occurrence, in an attempt to filter out the effect of dependencies that are not directly relevant for collocation extraction. In each case we show the effect of the extension in an evaluation against an extraction gold standard.

If time permits, I will end by discussing some other uses of association measures in (computational) corpus linguistics, in particular ongoing work on the effect of 'associatedness' on object fronting in Dutch.


Simon Dobnik (Oxford University): On learning the semantics of spatial words and generating extractive summaries of scientific papers

The talk describes two projects. In the first part of the talk I discuss how the semantics of spatial expressions can be grounded in the sensory data of a mobile robot. The mapping between the human conceptual domain and that used to represent space in physical sciences and mobile robotics is learned with machine learning. We discuss the performance classifiers to generate new contextually appropriate descriptions. In the second part of the talk I discuss my work with Maria Liakata (Aberystwyth University, EBI: Cambridge) on generating and evaluating extractive summaries of chemistry scientific papers. The sentences chosen for the summary are tagged for core scientific concepts such as Background, Hypothesis and Result which correspond to methods used in scientific investigation.


Richard Johansson (University of Trento): Semantic role extraction with applications

In this presentation, we give an introduction to semantic role
representations such as FrameNet, VerbNet, and PropBank, and their
potential benefits in practical NLP applications.  We describe how these
representations can be automatically extracted from free text, and examine
how this task is interrelated with syntactic parsing. In addition, we show
how parallel corpora can mitigate the problem of limited data availability
for development of semantic role extractors for small languages.

Date: 2011-03-21 13:30 - 15:50

Location: K332, Lennart Torstenssonsgatan 6

Permalink

add to Outlook/iCal

To the top

Page updated: 2011-03-17 17:11

Send as email
Print page
Show as pdf

X
Loading