Campus Event Calendar
Event Entry
What and Who
Knowledge-rich models for high-end NLP applications
Prof. Simone Paolo Ponzetto
Universität Mannheim
Talk
AG 5, MMCI
AG Audience
English
Note: We use this to send email in the morning.
Date, Time and Location
Thursday, 13 February 2014
11:00
60 Minutes
E1 4
024
Saarbrücken
Warning: New Room!
Abstract
The Web contains vast amounts of textual content which needs to be
automatically semantified (i.e. fully structured and annotated with
semantic information) in order to conform to the vision of a Web of
semantic data, and enable next-generation applications like, for
instance, semantic search. Semantic information, furthermore, is highly
intertwined with knowledge, since knowledge-rich methods have been shown
to achieve state-of-the-art performance on tasks that are essential for
generating semantic structure like word sense and entity disambiguation
and, conversely, semantified data can be used to further extend existing
repositories of machine-readable knowledge.
In this talk I will elaborate on this vision of a synergistic approach
to structured knowledge and semantic information by presenting an
overview of recent work on exploiting wide-coverage knowledge sources
for a variety of Natural Language Processing (NLP) tasks. We first
introduce methods to leverage multilingual knowledge from a very large
lexical database in order to achieve state-of-the-art performance on
different lexical understanding tasks, e.g., sense disambiguation and
word similarity. Next, we show how information from existing
wide-coverage ontologies, like YAGO or DBpedia, can be used to provide
structured (i.e., graph-based), semantically-rich representations of
texts, which can then be used to achieve robust performance on even more
complex tasks such as computing entity ranking and document similarity.
Our results confirm the notion that NLP applications can benefit
substantially from large amounts of knowledge to achieve human-level
performance on complex language processing tasks. Nevertheless, we argue
that much still remains to be done in terms of more sophisticated
modelling and depth of representation for both conceptual knowledge and
textual content.
automatically semantified (i.e. fully structured and annotated with
semantic information) in order to conform to the vision of a Web of
semantic data, and enable next-generation applications like, for
instance, semantic search. Semantic information, furthermore, is highly
intertwined with knowledge, since knowledge-rich methods have been shown
to achieve state-of-the-art performance on tasks that are essential for
generating semantic structure like word sense and entity disambiguation
and, conversely, semantified data can be used to further extend existing
repositories of machine-readable knowledge.
In this talk I will elaborate on this vision of a synergistic approach
to structured knowledge and semantic information by presenting an
overview of recent work on exploiting wide-coverage knowledge sources
for a variety of Natural Language Processing (NLP) tasks. We first
introduce methods to leverage multilingual knowledge from a very large
lexical database in order to achieve state-of-the-art performance on
different lexical understanding tasks, e.g., sense disambiguation and
word similarity. Next, we show how information from existing
wide-coverage ontologies, like YAGO or DBpedia, can be used to provide
structured (i.e., graph-based), semantically-rich representations of
texts, which can then be used to achieve robust performance on even more
complex tasks such as computing entity ranking and document similarity.
Our results confirm the notion that NLP applications can benefit
substantially from large amounts of knowledge to achieve human-level
performance on complex language processing tasks. Nevertheless, we argue
that much still remains to be done in terms of more sophisticated
modelling and depth of representation for both conceptual knowledge and
textual content.
Contact
Petra Schaaf
5000
--email hidden
passcode not visible
logged in users only
Petra Schaaf, 02/10/2014 13:55
Petra Schaaf, 02/10/2014 13:52 -- Created document.
Petra Schaaf, 02/10/2014 13:52 -- Created document.