DIMSUM: Discourse in Mathematical Reasoning as a Supervision Module
This addresses the issue of limited data or inferior training for mathematical reasoning in language models, though it is incremental as it builds on existing discourse analysis methods.
The paper tackles the problem of improving mathematical reasoning in language models by using discourse structure as a supervision module, showing that it boosts performance for models like Llama2 13b by up to 160% and enhances out-of-distribution predictions.
We look at reasoning on GSM8k, a dataset of short texts presenting primary school, math problems. We find, with Mirzadeh et al. (2024), that current LLM progress on the data set may not be explained by better reasoning but by exposure to a broader pretraining data distribution. We then introduce a novel information source for helping models with less data or inferior training reason better: discourse structure. We show that discourse structure improves performance for models like Llama2 13b by up to 160%. Even for models that have most likely memorized the data set, adding discourse structural information to the model still improves predictions and dramatically improves large model performance on out of distribution examples.