The Importance of Causal Language Modeling

Is it more akin to say like named entity recognition, but that task of assigning labels to each token. So you would say label this token as a cause token and an effect token. Is that it? Would that be like a more way to bridge the gap between NLP and causal inference? Yeah. You wouldn't want to have a model that looks both ways, right? But in terms of objects in a story, might be that the causes mentioned later in the sentence where the effect is mentioned earlier. That could make sense.