Manual annotation of evaluative language expressions : bridging discourse and corpus approaches
Author
Summary, in English
The analysis of evaluation in text poses significant methodological challenges, which mainly arise from the fact that: a) evaluative meanings can be expressed through an open-ended range of lexico-grammatical resources; b) they can span multiple words; c) context and co-text play a key role in determining the evaluative meaning of words or phrases; d) the interpretation of evaluation in text depends on the reader’s/analyst’s reading position and is, therefore, necessarily subjective.
These complexities have so far seriously limited the development of corpus-driven description of this phenomenon. This study addresses these methodological issues through the application of multivariate usage-feature / profile-based analysis (Geeraerts et al., 1994; Gries, 2003). This method bridges quantitative and qualitative perspectives and has been successfully applied to the description of various semantic and morphosyntactic phe- nomena. The method relies on the use of manual annotation software and is based on explicit criteria for the identification of evaluative items, including inter-annotator agreement metrics to control for rater bias. The results of the annotation are modelled with mutlivariate statistics. This step identifies structural patterns in the data and tests the accuracy of the description with predictive modelling.
These complexities have so far seriously limited the development of corpus-driven description of this phenomenon. This study addresses these methodological issues through the application of multivariate usage-feature / profile-based analysis (Geeraerts et al., 1994; Gries, 2003). This method bridges quantitative and qualitative perspectives and has been successfully applied to the description of various semantic and morphosyntactic phe- nomena. The method relies on the use of manual annotation software and is based on explicit criteria for the identification of evaluative items, including inter-annotator agreement metrics to control for rater bias. The results of the annotation are modelled with mutlivariate statistics. This step identifies structural patterns in the data and tests the accuracy of the description with predictive modelling.
Department/s
Publishing year
2013
Language
English
Full text
- Available as PDF - 71 kB
- Download statistics
Document type
Conference paper: abstract
Topic
- Languages and Literature
Conference name
Evaluative Language and Corpus Linguistics Workshop - Corpus Linguistics Conference
Conference date
2013-07-22 - 2013-07-26
Conference place
Lancaster, United Kingdom
Status
Published