The aim of this project
is the integration of
lexicogrammatical
features of language
production, as modeled
in the Greek Foreign
Language Learner Profile
Project, in an algorithm
for the automated
assessment of language
performance. A
significant gap in
existing research in the
field of automatic free
text marking is related
to the lack of systems
grading linguistic
instantiations in
specific communicative
contexts. In this
project, we design and
develop a novel system
that will classify KPG
scripts into grading
bands for each
proficiency level, on
the basis of linguistic
properties and their
function in the
communicative situation
to which the text
pertains. The system
will classify ungraded
scripts by computing
their similarities with
a gold corpus of graded
scripts. We will
experiment with various
combinations of features
(syntactic-lexical,
lexical-semantic,
syntactic-semantic, all
three) producing
different versions of
the model, which will be
evaluated against the
performance of related
models. The system will
be piloted in the KPG
exam battery and a
detailed technical study
will be carried out
reporting on the
reliability of the
assessment in relation
to the reduction of the
human resources cost for
script grading.
