Lune

EMNLP Findings2023

Towards Concept-Aware Large Language Models

Chen Shani, Jilles Vreeken, Dafna Shahaf

Abstract

Concepts play a pivotal role in various human cognitive functions, including learning, reasoning and communication. However, there is very little work on endowing machines with the ability to form and reason with concepts. In particular, state-of-the-art large language models (LLMs) work at the level of tokens, not concepts.

In this work, we analyze how well contemporary LLMs capture human concepts and their structure. We then discuss ways to develop concept-aware LLMs, taking place at different stages of the pipeline. We sketch a method for pretraining LLMs using concepts, and also explore the simpler approach that uses the output of existing LLMs. Despite its simplicity, our proof-of-concept is shown to better match human intuition, as well as improve the robustness of predictions. These preliminary results underscore the promise of concept-aware LLMs.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

3 top-tier papers have cited it since. Know what they found before your reviewers do.

Questions to start from

Lune can draw on

the full paper · the top-tier papers that cite it · related work

Get the answer

Free to start. No credit card required.

Lineage

  • Knowledge Infused Decoding cites SenseBERT: Driving Some Sense into BERT
  • Toolformer: Language Models Can Teach Themselves to Use Tools cites PEER: A Collaborative Language Model
  • Toolformer: Language Models Can Teach Themselves to Use Tools cites HTLM: Hyper-Text Pre-Training and Prompting of Language Models
Builds onCited byLeft to right by year; the more cited, the higher.

Cited by top-tier papers3

  1. From Tokens to Thoughts: How LLMs and Humans Trade Compression for MeaningChen Shani, Liron Soffer, Dan Jurafsky, Yann LeCun et al.ICLR 2026 · 38 citations
  2. Vision-and-Language Training Helps Deploy Taxonomic Knowledge but Does Not Fundamentally Alter ItYulu Qin, Dheeraj Varghese, Adam Dahlgren Lindström, Lucia Donatelli et al.NeurIPS 2025 · 11 citations
  3. Understanding Subword Compositionality of Large Language ModelsQiwei Peng, Yekun Chai, Anders SøgaardEMNLP 2025

Builds on8

  1. Toolformer: Language Models Can Teach Themselves to Use ToolsTimo Schick, Jane Dwivedi-Yu, Roberto Dessì, Roberta Raileanu et al.NeurIPS 2023 · 5,989 citations
  2. HTLM: Hyper-Text Pre-Training and Prompting of Language ModelsArmen Aghajanyan, Dmytro Okhonko, Mike Lewis, Mandar Joshi et al.ICLR 2022 · 82 citations
  3. PEER: A Collaborative Language ModelTimo Schick, Jane A. Yu, Zhengbao Jiang, Fabio Petroni et al.ICLR 2023 · 44 citations
  4. SenseBERT: Driving Some Sense into BERTYoav Levine, Barak Lenz, Or Dagan, Ori Ram et al.ACL 2020 · 27 citations
  5. Knowledge Infused DecodingRuibo Liu, Guoqing Zheng, Shashank Gupta, Radhika Gaonkar et al.ICLR 2022 · 18 citations

Related papers

  1. Conceptual structure coheres in human cognition but not in large language modelsSiddharth Suresh, Kushin Mukherjee, Xizheng Yu, Wei-Chun Huang et al.EMNLP 2023 · 7 citations
  2. The Strawberry Problem: Emergence of Character-level Understanding in Tokenized Language ModelsAdrian Cosma, Stefan Ruseti, Emilian Radoi, Mihai DascaluEMNLP 2025 · 13 citations
  3. Agentic-ToM: Cognition-Inspired Agentic Processing For Enhancing Theory of Mind ReasoningSneheel Sarangi, Chetan Talele, Hanan SalamEMNLP Findings 2025 · 2 citations
  4. Concept-aware Data Construction Improves In-context Learning of Language ModelsMichal Stefánik, Marek Kadlcík, Petr SojkaACL Findings 2024 · 1 citation
  5. What is a protest anyway? Codebook conceptualization is still a first-order concern in LLM-era classificationAndrew Halterman, Katherine A. KeithACL 2026 · 3 citations
Dusk over the sea between two cliffs drawn in fine vertical lines