Automatic framenet-guided extraction of terminology: an application to the cord-19 electronic corpus

ABSTRACT

The evolution of Terminology is joined to new technologies and the development of work platforms or interfaces that allow creating a technical glossary semi-automatically or even automatically. Terms and their properties are connected formally to the expression of knowledge of specialized fields in which they occur, so automatic approaches are not only faced with the task of determining which are the terminological units of a given field, but to express how such information is structured in their technical field. Most of the terms occurring in a scientific domain are also found in other disciplines and even in everyday language. Terms are often present on the lexical stock of languages and share with lexical units a complex set of relationships. Frame Semantics is a particularly attractive model for the terminological work, interested in accounting for the connection between the conceptual structure of a specialized field and the elements used to transmit this knowledge. This has led to many researchers to use FrameNet as a way of representing terminology. FrameNet is an online resource for English based on Frame Semantics and supported by corpus evidence. A frame is founded on the basis that certain words evoke certain situations in which particular participants take place. These situations or frames are stereotyped structures representing areas of sociocultural experience and knowledge. We present a statistical approach based on corpus able to select most representative FrameNet frames that best represent a set of electronic texts on COVID-19 and show which of their lexical units work as terminological units. Results confirm that this methodology can be a good support for terminographic work, since it not only allows the extraction of terminological units, but also the use of the FrameNet framework to structure this knowledge.