Word Meanings in Transformer Language Models

TL;DR

This paper explores how transformer language models encode word meanings by clustering token embeddings and analyzing their sensitivity to semantic and psycholinguistic features, revealing rich semantic information in the embeddings.

Contribution

It demonstrates that transformer models encode diverse semantic information in token embeddings, challenging meaning eliminativist hypotheses.

Findings

Token embedding clusters are sensitive to semantic features

Transformer models encode a wide range of psycholinguistic information

Results challenge theories that deny semantic content in LLM representations

Abstract

We investigate how word meanings are represented in the transformer language models. Specifically, we focus on whether transformer models employ something analogous to a lexical store - where each word has an entry that contains semantic information. To do this, we extracted the token embedding space of RoBERTa-base and k-means clustered it into 200 clusters. In our first study, we then manually inspected the resultant clusters to consider whether they are sensitive to semantic information. In our second study, we tested whether the clusters are sensitive to five psycholinguistic measures: valence, concreteness, iconicity, taboo, and age of acquisition. Overall, our findings were very positive - there is a wide variety of semantic information encoded within the token embedding space. This serves to rule out certain "meaning eliminativist" hypotheses about how transformer LLMs process semantic information.