Parallel Dependency Treebank Annotated with Interlinked Verbal Synonym Classes and Roles

Abstract

We present an ongoing project of enriching an annotation of a parallel dependency treebank, namely the Prague Czech-English Dependency Treebank, with verb-centered semantic annotation using a bilingual synonym verb class lexicon, CzEngClass. This lexicon, in turn, links the predicate occurrences in the corpus to various external lexicons, such as FrameNet, VerbNet, PropBank frame files, OntoNotes, and WordNet. We briefly describe the content of the CzEngClass synonym class lexicon and then we focus on its use for an enrichment of corpus annotation, which proceeds in two steps-automatic preprocessing and manual correction. This paper describes a first milestone of a long-term project; so far, approx. 100 CzEngClass classes, containing about 1800 different verbs each for both Czech and English, are available for such annotation. The corpus coverage at the moment is about 50%, allowing us to extract some basic statistics and discover a set of issues that appeared during the annotation process. The ultimate goal is to have a high-coverage, multilingual verbal synonym lexicon and corpora with all events annotated by such lexicon, to serve both theoretical studies in lexical semantic, translatology, corpus annotation studies etc. as well as a usable resource for training automatic semantic text processing systems for event/participant detection and linking and for general information extraction.

A model for representing the semantics of MWEs: From lexical semantics to the semantic annotation of complex predicates

“…A few years later, the Brandeis Semantic Ontology (BSO) seeks to extend the English SIMPLE lexicon (Pustejovsky et al, 2006 ). At the syntax-semantics interface, SynSemClass (Urešová et al, 2018a , b ), is a bilingual synonym lexicon organized on the basis of contextually based synonymy and valency of verbs in a bilingual setting; at the heart of the bilingual lexicon lays the analysis of semantic “equivalence” (synonymy or near synonymy) of verb senses, and their valency behavior in parallel Czech-English language resources. In this respect, semantic MWE-aware lexicons, i.e., WordNet (Fellbaum, 1998 ), Verbnet (Kipper et al, 2008 ), SAID (Kuiper et al, 2003 ), and WikiMwe (Hartmann et al, 2012 ) give an account of various types of MWEs—yet they are solely focused on their semantic representation overlooking other aspects.…”

Spanish Synonyms as Part of a Multilingual Event-Type Ontology

“…Previous work on the extension of the ontology described a minimal set of resources required for the addition of new languages (Urešová et al 2022). In particular, the necessary resources required are two: a parallel corpus and (at least) one lexical resource with information on verbal valency.…”

“…Any discrepancies that arose during the annotation process were discussed as needed. For the task of annotation, which involves mapping roles to arguments, identifying external links, and selecting examples, we used the SynEd editor (Urešová et al 2018a;Fučíková et al 2023), a tool specifically designed for this purpose by the SynSemClass maintainers and refactored to adapt it to any number of languages.…”