In the framework of perspectivism, analyzing how people perceive pragmatic phenomena, like irony, is relevant for deeply understanding the different points of view, and for creating more robust perspective-aware models. This paper presents a linguistic analysis of irony perception in 11 perspectivist models. Each model is trained on annotations by crowd-sourcing workers different in gender, age, and nationalities. Due to the sparsity of the dataset, we examine the texts classified as ironic and not-ironic by these perspectivist models, and identify linguistic patterns that all perspectives associate with irony. To our knowledge, we are the first to also provide evidence for the different linguistic patterns perceived as ironic by a specific perspective. For example, models trained on data annotated by American and Australian annotators are more inclined to classify a text as ironic when it includes a negative sentiment, while models trained on data annotated by the youngest annotators are particularly influenced by words related to immoral behaviors. Warning: This paper could contain content that is offensive or upsetting for the reader.
Does Anyone see the Irony here? Analysis of Perspective-aware Model Predictions in Irony Detection
Frenda S.;Lo S. M.;Casola S.;Basile V.;
2023-01-01
Abstract
In the framework of perspectivism, analyzing how people perceive pragmatic phenomena, like irony, is relevant for deeply understanding the different points of view, and for creating more robust perspective-aware models. This paper presents a linguistic analysis of irony perception in 11 perspectivist models. Each model is trained on annotations by crowd-sourcing workers different in gender, age, and nationalities. Due to the sparsity of the dataset, we examine the texts classified as ironic and not-ironic by these perspectivist models, and identify linguistic patterns that all perspectives associate with irony. To our knowledge, we are the first to also provide evidence for the different linguistic patterns perceived as ironic by a specific perspective. For example, models trained on data annotated by American and Australian annotators are more inclined to classify a text as ironic when it includes a negative sentiment, while models trained on data annotated by the youngest annotators are particularly influenced by words related to immoral behaviors. Warning: This paper could contain content that is offensive or upsetting for the reader.File | Dimensione | Formato | |
---|---|---|---|
C17_2023.pdf
Accesso aperto
Tipo di file:
PDF EDITORIALE
Dimensione
307.89 kB
Formato
Adobe PDF
|
307.89 kB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.