Aims and objectives

WP 1: Correcting VMWE annotations in the PMB

The objective is to create a dataset that includes full semantic representations of sentences containing verbal multiword expressions (VMWEs) in order to train and evaluate semantic parsers wrt. such expressions. To this end, we will extend part of the Parallel Meaning Bank (PMB) for English and German in an annotation effort focused on potential VMWEs. VMWEs are one area where the annotation guidelines of the PMB are not yet fully worked out and as a consequence, annotation is incomplete, so this effort will not only benefit our project but also the PMB.

WP 2: MWE-aware syntactic parsing

The goal of WP 2 is to propose a parsing architecture which will allow to handle MWE identification jointly with syntactic parsing. We will rely on a particular syntactic framework, TAG, which should facilitate further semantic analysis thanks to the syntax/semantics interface, keeping in mind that the resulting architecture should also allow the use of other syntactic frameworks.

WP 3: MWE-aware semantic parsing

WP 3 is concerned with the development of semantic parsers that are capable of correctly interpreting MWEs and distinguishing between idiomatic MWEs and literal uses. The focus is on verbal MWEs but other classes might be included as well. We will mainly focus on parsing to DRSs, using the PMB further enriched with semantic VMWE annotations (see WP 1) and with TAG ETs and derivation trees.