Decoder-based Sense Knowledge Distillation

TL;DR

This paper introduces DSKD, a novel framework that incorporates lexical sense knowledge into decoder-based language models during training, improving their semantic understanding without increasing inference complexity.

Contribution

The paper presents a new method for integrating lexical sense knowledge into decoder models, addressing a gap in prior work focused on encoder models.

Findings

DSKD improves knowledge distillation performance for decoder models.

Decoders trained with DSKD better capture structured semantic information.

The approach maintains training efficiency without requiring dictionary lookups during inference.

Abstract

Large language models (LLMs) learn contextual embeddings that capture rich semantic information, yet they often overlook structured lexical knowledge such as word senses and relationships. Prior work has shown that incorporating sense dictionaries can improve knowledge distillation for encoder models, but their application to decoder as generative models remains challenging. In this paper, we introduce Decoder-based Sense Knowledge Distillation (DSKD), a framework that integrates lexical resources into the training of decoder-style LLMs without requiring dictionary lookup at inference time. Extensive experiments on diverse benchmarks demonstrate that DSKD significantly enhances knowledge distillation performance for decoders, enabling generative models to inherit structured semantics while maintaining efficient training.