C58 : the Greek Discourse Relations Corpus
WHO. WE.
ARE.
C
5
8
project
A FEW WORDS ABOUT US...
Project leader
-
Alex Tantos (assistant professor in Computational and Corpus Linguistics at Aristotle University of Thessaloniki)
Members
-
George Chaziioanidis (Linguist)
-
Katerina Lykou (Linguist)
-
Meropi Papatheochari (Linguist)
-
Αntonia Samara (Linguist)
-
Konstantinos Vlachos (Linguist)
A FEW WORDS ABOUT THE PROJECT...
C58 is built as part of a one-year project funded by Aristotle University of Thessaloniki starting on September of 2014. The project aimed at two things: a) the compilation of an annotated corpus of Discourse Relations for Greek and b) implementing a classifier for predicting discourse relations between the current utterance and the previous discourse. In its first version,
C58 consists of 58 annotated texts (found here), sampled from a general purpose corpus of Modern Greek compiled by the Center of Greek Language in Thessaloniki.
The texts belong to various journalistic genres compiled from two of the main newspapers in Greece, "Ta Nea" and "Makedonia". More info about the C58's implementation and theoretical background can be found on what we do and papers' section.
A FEW WORDS ABOUT OUR TOOLS...
-
Python is a flexible scripting language used for various text preprocessing tasks, such as sampling of texts, getting and cleaning the data and, therefore, suitable for our purposes.
-
We used python for a number of processing steps and will alsouse it for creating discourse graphs in order to depict discourse dependencies resembling SDRT's graphs.
BRAT
-
brat is an open source web-based annotation tool, developed by the University of Tokyo, and is especially helpful for annotating discourse relations between textual segments and has a set of interesting features that ease the annotation effort, such as collaborative annotation via a web server.
VERBNET [VN]
-
The largest on-line verb lexicon currently available for English that served as the basis for our intrasentential annotation.
-
Verbnet is organized into verb classes extending Levin (1993) classes and integrating Framenet classes. Each verb class in VN includes information on thematic roles, arguments' selectional restrictions and their syntactic and semantic description. Our annotation scheme is partly based on VN and adjusted for Greek.