Numeric data cannot be accurately entailed due to the limitations of the language model #33
Description
ProVe uses pre-trained language model for text entailment. As known well, the language model is inappropriate to analyze numbers. For example, ProVe would fail to check supportiveness for death or birth dates.
Is there any good idea for this?