# Reversing Arrows in Large Language Models

arXiv: https://arxiv.org/abs/2608.03512 · HTML: https://mlsanity.com/paper/2608.03512/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 12 of its 18 references that are in the map.
- First posted on arXiv 4 Aug 2026.

## Builds on (most cited first, 12 of 12)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Mistral 7B](https://mlsanity.com/paper/2310.06825/) (2023)
- [LaMDA: Language Models for Dialog Applications](https://mlsanity.com/paper/2201.08239/) (2022)
- [FewRel: A Large-Scale Supervised Few-Shot Relation Classification Dataset with State-of-the-Art Evaluation](https://mlsanity.com/paper/1810.10147/) (2018)
- [Text2KGBench: A Benchmark for Ontology-Driven Knowledge Graph Generation from Text](https://mlsanity.com/paper/2308.02357/) (2023)
- [Ontology Generation using Large Language Models](https://mlsanity.com/paper/2503.05388/) (2025)
- [An Investigation of LLMs' Inefficacy in Understanding Converse Relations](https://mlsanity.com/paper/2310.05163/) (2023)
- [KERMIT: Knowledge Graph Completion of Enhanced Relation Modeling with Inverse Transformation](https://mlsanity.com/paper/2309.14770/) (2023)
- [Function-words Enhanced Attention Networks for Few-Shot Inverse Relation Classification](https://mlsanity.com/paper/2204.12111/) (2022)
- [Taxonomical hierarchy of canonicalized relations from multiple Knowledge Bases](https://mlsanity.com/paper/1909.06249/) (2019)
- [Do Models Learn the Directionality of Relations? A New Evaluation: Relation Direction Recognition](https://mlsanity.com/paper/2105.09045/) (2021)
- [Knowledge Prompting: How Knowledge Engineers Use Large Language Models](https://mlsanity.com/paper/2408.08878/) (2024)

## Abstract (arXiv)

Large language models (LLMs) have achieved strong performance on text-to-knowledge graph generation and related tasks. Nevertheless, it is still unclear whether they accurately model the direction-dependent semantics of inverse relations, in which reversing the order of the arguments alters the meaning of a relation (e.g., textitmother versus textitchild). To the best of our knowledge, this work presents the first systematic study of inverse relation directionality in LLMs, using a benchmark consisting of 5,457 instances spanning 27 distinct inverse relation labels. We evaluate five open-source LLMs under a multiple-choice prompting framework and further examine the influence of relation descriptions and entity representations by substituting the original entities with synthetic and masked entities. Our findings reveal systematic asymmetries in inverse relation classification across LLMs, indicate that relation descriptions do not consistently improve performance, and show that model performance can be sensitive to variations in entity representations.