PARSEME

PARSEME is a European research initiative focused on multiword expressions (MWEs) in natural language processing. It operates as a network of excellence funded by the European Commission to advance the study and processing of MWEs across languages and applications. MWEs are sequences of words whose overall meaning cannot be deduced from the meanings of their parts, making their recognition and interpretation a challenge for NLP systems, machine translation, parsing, and information extraction. PARSEME aims to improve linguistic theory and computational methods for MWEs by coordinating research, sharing data, and producing practical resources.

The initiative develops and promotes standard annotation guidelines for identifying MWEs, along with multilingual corpora and lexicons annotated with MWE information. It supports evaluation frameworks to benchmark MWE recognition, disambiguation, and integration with NLP pipelines. A core activity is the organization of shared tasks and workshops that provide common tasks, data sets, and evaluation metrics to compare approaches across languages.

PARSEME also curates publicly available resources, such as annotated corpora and reference materials, to facilitate reproducibility and further research.

The influence of PARSEME lies in its contribution to standardizing MWE annotation practices, aligning researchers across Europe and beyond, and enriching NLP through improved handling of MWEs in parsing, translation, and semantic processing. After its period of active coordination, many resources remain accessible through the PARSEME project portals and affiliated repositories, continuing to support ongoing work on MWEs.