Exploring lexical factors in semantic annotation: insights from the classification of nouns in French
Université de Fribourg BLE-BLL
- Barque, Lucie ORCID Universit´e Sorbonne Paris Nord
- Huyghe, Richard ORCID University of Fribourg
- Foegel, Martial Laboratoire de linguistique formelle, CNRS
- 2025
Published in:
- Language Resources and Evaluation. - Springer Science and Business Media LLC. - 2025, vol. 59, no. 3, p. 2143-2167
semantic classification
inter-annotator agreement
supersenses
lexical ambiguity
word frequency
concreteness
semantic hybridity
English
This paper investigates how different lexical factors influence inter-annotator
agreement in a semantic annotation task, with the level of agreement serving
as an indicator of task complexity. The study uses a dataset of approximately
5,000 corpus instances of French nouns, each double annotated with supersenses
representing broad semantic classes such as Person, Object and Event. Through
statistical analysis, the study evaluates the individual impact of word ambiguity,
frequency, concreteness and meaning hybridity on semantic classification. The
results show that the four lexical factors under study are correlated with interannotator agreement and that they jointly predict a significant portion of the
agreement data. More importantly, they contribute to a better understanding of
how speakers categorise words semantically, while allowing for a comparison of
manual and automated semantic classification based on lexical properties
agreement in a semantic annotation task, with the level of agreement serving
as an indicator of task complexity. The study uses a dataset of approximately
5,000 corpus instances of French nouns, each double annotated with supersenses
representing broad semantic classes such as Person, Object and Event. Through
statistical analysis, the study evaluates the individual impact of word ambiguity,
frequency, concreteness and meaning hybridity on semantic classification. The
results show that the four lexical factors under study are correlated with interannotator agreement and that they jointly predict a significant portion of the
agreement data. More importantly, they contribute to a better understanding of
how speakers categorise words semantically, while allowing for a comparison of
manual and automated semantic classification based on lexical properties
- Faculty
- Faculté des lettres et des sciences humaines
- Department
- Département de français
- Language
-
- English
- License
- Rights reserved
- Open access status
- green
- Identifiers
-
- DOI 10.1007/s10579-024-09801-4
- ISSN 1574-020X
- Persistent URL
- https://folia.unifr.ch/global/documents/332754
Statistics
Document views: 18
File downloads:
- iaa_paper_final.pdf: 10