Humboldt-Universität zu Berlin - Mathematisch-Naturwissenschaftliche Fakultät - Institut für Informatik

Liebe Institutsangehörige,

am Dienstag, den 19.12.2023 um 9 s.t. stellt Michel Schwab
die wesentlichen Ergebnisse seiner Forschung zum Thema *Detection, Extraction and Analysis of the stylistic device Vossian Antonomasia in large text corpora using machine learning* vor (Abstract anbei).

Der Vortrag findet im Humboldt Kabinett, RUD 25 statt.

Der angekündigte Vortrag wird kurzfristig online stattfinden: Eine Zoom-Einladung finden Sie hier. (nur mit Informatik-Account)

Alle Interessierten sind herzlich eingeladen.


Viele Grüße

Robert Jäschke

Abstract:
We focus on the automated detection, extraction and analysis of
Vossian Antonomasia (VA), a rhetorical device that employs named
entities to transfer attributes from a source to a target entity. The
first automated methods we introduce are initially based on syntactic
patterns. They employ neural network-based classifiers, which
significantly outperform rule-based methods for the general detection
of VA. These neural models utilize both non-contextual and contextual
word embeddings, long short-term memory networks, and pre-trained
language models. We also present methods for chunk extraction by
modeling the problem as a sequence tagging task and we evaluate the
performance on real-world datasets. Additionally, we explore
cross-lingual extraction models and develop methods for extracting
target entities in entire texts. For a deeper understanding of VA, we
employ clustering algorithms, dimensionality reduction techniques, and
lexical topic modeling to explore the connections between the
different chunks of VA expressions.