# CoastTerm: a Corpus for Multidisciplinary Term Extraction in Coastal Scientific Literature

**Authors:** Julien Delaunay, Hanh Thi Hong Tran, Carlos-Emiliano González-Gallardo, Georgeta Bordea, Mathilde Ducos, Nicolas Sidere, Antoine Doucet, Senja Pollak, Olivier De Viron  
**arXiv:** [2406.09128](https://arxiv.org/abs/2406.09128)  
**Date:** 2024-06-13  
**Summary page:** https://www.summarizepaper.com/en/paper/2406.09128v1/

## Key points

- Collaboration among diverse stakeholders and disciplines is essential for formulating effective environmental protection policies in coastal areas impacted by climate change.
- A specialized corpus of scientific abstracts related to coastal areas was used for Automatic Term Extraction (ATE) and Classification (ATC) tasks.
- Domain terms and their roles in coastal systems were automatically extracted using monolingual and multilingual transformer models inspired by the ARDI framework.
- An annotation process involving students specializing in Earth Sciences, a domain expert, and a PhD student was conducted to annotate abstracts with a moderate agreement level of 43% over two months.
- Three domain-relevant knowledge bases were utilized to pre-annotate terms in a subset of abstracts, leading to refined term boundaries and the creation of datasets for studying coastal areas.

## Summary

, , , , 


In an effort to address the growing impact of climate change on coastal areas, particularly in active but fragile regions, collaboration among diverse stakeholders and disciplines is essential to formulate effective environmental protection policies. To aid in this endeavor, a specialized corpus comprising 2,491 sentences from 410 scientific abstracts related to coastal areas was introduced for the purpose of Automatic Term Extraction (ATE) and Classification (ATC) tasks. Drawing inspiration from the ARDI framework, which focuses on identifying Actors, Resources, Dynamics, and Interactions within a system, domain terms and their roles in coastal systems were automatically extracted using monolingual and multilingual transformer models. The annotation process involved two students specializing in Earth Sciences, a domain expert, and a PhD student who all simultaneously annotated papers using the INCEpTION tool. Over two months, 215 abstracts were annotated with a moderate agreement level of 43%, indicating the difficulty of manual annotation. Additionally, three domain-relevant knowledge bases were utilized to pre-annotate terms in a subset of abstracts, leading to refined term boundaries. Two datasets were created for studying coastal areas by homogenizing manually annotated subsets with KB-recommended annotations. The datasets underwent manual adaptation for terminology extraction by removing relations and pronouns indicating coreferences. The CoastTerm corpus for term extraction was developed from a larger collection of papers from Scopus containing terms such as "coastal areas" or "littoral" in their titles or abstracts. The annotation process involved undergraduate Master's students specialized in Earth Sciences conducting joint entity and relation extraction tasks based on guidelines provided by PhD students and domain experts. Only sentences providing information on the functioning of coastal zones were annotated using labels such as "Actor" for stakeholders and "Resource" for goods utilized within the system. Overall, this study represents an important step towards developing a specialized Knowledge Base dedicated to coastal areas by leveraging automated term extraction methods and incorporating insights from manual annotation processes.