Performance analysis of SOTA Transformer Network in Numerical Expression Calculation

Computing Numerical expressions by analyzing the sentences is an intricate task in the domain of Natural Language Processing (NLP). Moreover, the noisy ("in-word") and erroneous representation of numerical expressions make the task more complex. In this work we have studied the performance of Transformer network models for solving numerical expressions. We have reported performance measurements for the task both in the word-level and character-level transformers. Observations reveal that though the transformer networks perform outstanding in sequence understanding tasks, the models under-perform on more logical calculative numerical evaluations.