Mitigating Reversal Curse in Large Language Models via Semantic-aware Permutation Training
This addresses a fundamental reasoning limitation in LLMs that hinders progress toward AGI, though it is an incremental improvement over prior permutation methods.
The paper tackles the reversal curse in large language models, where models fail at bidirectional reasoning (e.g., knowing 'A's father is B' but not 'B's child is A'), by proposing Semantic-aware Permutation Training (SPT), which segments sentences into semantic units and permutes them during training, resulting in performance on reversed questions approximating that on forward ones and significant improvements over existing works.
While large language models (LLMs) have achieved impressive performance across diverse tasks, recent studies showcase that causal LLMs suffer from the "reversal curse". It is a typical example that the model knows "A's father is B", but is unable to reason "B's child is A". This limitation poses a challenge to the advancement of artificial general intelligence (AGI), as it suggests a gap in the models' ability to comprehend and apply bidirectional reasoning. In this paper, we first conduct substantial evaluation and identify that the root cause of the reversal curse lies in the different word order between the training and inference stage, namely, the poor ability of causal language models to predict antecedent words within the training data. Accordingly, permutation on the training data is considered as a potential solution, since this can make the model predict antecedent words or tokens. However, previous permutation methods may disrupt complete phrases or entities, thereby posing challenges for the model to comprehend and learn from training data. To address this issue, we propose Semantic-aware Permutation Training (SPT), which addresses this issue by segmenting the training sentences into semantic units (i.e., entities or phrases) with an assistant language model and permuting these units before feeding into the model. Extensive experiments demonstrate that SPT effectively mitigates the reversal curse since the performance on reversed questions approximates that on the forward ones, and significantly advances the performance of existing works.