© 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.The state-of-the-art NLP models’ success advanced significantly as their complexity increased in recent years. However, these models tend to consider the statistical correlation between features which may lead to bias. Therefore, to build robust systems, causality should be considered while estimating the given task’s data generating process. In this study, we explore text-based causal inference on the irony and sarcasm detection problem. Additionally, we model the latent confounders by performing unsupervised data analysis, particularly clustering and topic modeling. The obtained results also provide insight for the causal explainability in irony detection.