Search papers, labs, and topics across Lattice.
The paper introduces a method for automatically augmenting commonsense knowledge graphs with negated relations to improve LLMs' understanding of negation. They generate two new corpora with over 2M negated if-then triples by applying a set of transformation rules to existing commonsense knowledge. Pre-training LLMs on these corpora demonstrates improved performance on downstream negation understanding tasks.
LLMs struggle with negated commonsense reasoning, but pre-training on automatically-generated negated knowledge triples can significantly improve their performance.
Negation is a common and important semantic feature in natural language, yet Large Language Models (LLMs) struggle when negation is involved in natural language understanding tasks. Commonsense knowledge, on the other hand, despite being a well-studied topic, lacks investigations involving negation. In this work, we show that commonsense knowledge with negation is challenging for models to understand. We present a novel approach to automatically augment existing commonsense knowledge corpora with negation, yielding two new corpora containing over 2M triples with if-then relations. In addition, pre-training LLMs on our corpora benefits negation understanding.
