Polar Gender Reference in Newspaper Texts

We have trained a classifier for detecting gender denoting nouns (noun phrases) in texts. On the basis of a sentiment inference system and the gender classifier, gender-tailored profiling in newspaper texts is realized. We discovered an imbalance wrt. gender denoting nouns and the role they take as sources or targets of verbs denoting positive or negative  relationships. 
Our goal was to get empirical access to the perception of gender, their roles and their reciprocal relations as portrayed in the news. Our empirical findings are based on  statistical hypothesis testing.

see (available from Zora):

Klenner, Manfred; Goehring, Anne; Yong-Ju Kim, Alison; Massey, Dylan  (2023). Gender-tailored Semantic Role Profiling for German. In: 12th International Conference on Computational Semantics (IWCS), Nancy, France, 20 Juni 2023 - 23 Juni 2023.

and:

Klenner, Manfred  (2023). Sentiment Inference for Gender Profiling. In: Language, Data, and Knowledge, Vienna, Austria, 12 September 2023 - 15 September 2023.

We investigate how gender authorship influences polar gender reference, i.e. whether reference is positive or negative. Given German language newspaper texts where the full name of the authors are known and their gender can be infered on the basis of the first name, nouns in the text with gender reference are classified on the basis of a gender classifier as female or male denoting nouns. If these nouns (and their phrases) carry a polar load, then they count towards the gender-specific statistics we are interested in. 
A polar load is either given via sentiment composition, if the noun itself is polar or is modified by a polar adjective, for instance. But also a verb-based analysis of the polar role a noun plays is carried out: is it framed by the verb as a positive or negative actor, is it framed as receiving a positive or negative effect.
Finally,  frequency of reported polar gender-gender relations (in favour, against) might depend on the gender of the author of the text and are considered in our empirical setting as well. Statistical hypothesis testing is carried out in order to find out whether significant gender-wise correlations exist. 

see:

Klenner, Manfred; Massey, Dylan  (2024). Is Gender Reference Gender-specific? Studies in a Polar Domain. (to appear in: Proceedings of the LREC-COLING 2024)