The Strawberry Problem: Emergence of Character-level Understanding in Tokenized Language Models

• URL: http://arxiv.org/abs/2505.14172v3
• Date: Mon, 15 Sep 2025 18:36:32 GMT
• Title: The Strawberry Problem: Emergence of Character-level Understanding in Tokenized Language Models
• Abstract summary: Large Language Models (LLMs) consistently fail at simple character-level tasks, such as counting letters in words, due to a fundamental limitation: tokenization.<n>We propose a lightweight architectural modification that significantly improves character-level reasoning while preserving the inductive advantages of subword models.<n>Our results bridge low-level perceptual gaps in tokenized LMs and provide a principled framework for understanding and mitigating their structural blind spots.
• Score: 2.531076482407163
• License: http://creativecommons.org/licenses/by-nc-nd/4.0/
• Abstract: Despite their remarkable progress across diverse domains, Large Language Models (LLMs) consistently fail at simple character-level tasks, such as counting letters in words, due to a fundamental limitation: tokenization. In this work, we frame this limitation as a problem of low mutual information and analyze it in terms of concept emergence. Using a suite of 19 synthetic tasks that isolate character-level reasoning in a controlled setting, we show that such capabilities emerge suddenly and only late in training. We find that percolation-based models of concept emergence explain these patterns, suggesting that learning character composition is not fundamentally different from learning commonsense knowledge. To address this bottleneck, we propose a lightweight architectural modification that significantly improves character-level reasoning while preserving the inductive advantages of subword models. Together, our results bridge low-level perceptual gaps in tokenized LMs and provide a principled framework for understanding and mitigating their structural blind spots. We make our code publicly available.

Related papers

• Context Staircase: Signature-Aligned Dynamics of Token Embeddings under Small Initialization [7.898646648931469]
  Token embeddings are the basic representational units that connect discrete tokens with continuous computation in language models.<n>We identify that the evolving embedding structures are closely related to token-conditioned label and contextual distributions.<n>We show that these embedding structures play an important role in both task learning and the incorporation of semantic structure into the embedding space.
  arXiv  Detail & Related papers  (2026-08-31T06:33:09Z)
• SLASH the Sink: Sharpening Structural Attention Inside LLMs [55.75072688988806]
  Large Language Models (LLMs) show remarkable semantic understanding but struggle with structural understanding when processing graph topologies in a serialized format.<n>We propose StructuraL Attention SHarpening (SLASH), which amplifies this internal structural understanding via a plug-and-play attention redistribution.
  arXiv  Detail & Related papers  (2026-05-11T12:59:07Z)
• Decoupling the Benefits of Subword Tokenization for Language Model Training via Byte-level Simulation [6.240815206375064]
  Subword tokenization is an essential part of modern large language models (LLMs)<n>In this work, we decouple the effects of subword tokenization by isolating them within a controlled byte-level pretraining pipeline.
  arXiv  Detail & Related papers  (2026-04-29T23:29:08Z)
• From Early Encoding to Late Suppression: Interpreting LLMs on Character Counting Tasks [49.57538588967748]
  Large language models (LLMs) exhibit failures on elementary symbolic tasks such as character counting in a word, despite excelling on complex benchmarks.<n>We uncover a consistent phenomenon across modern architectures, including LLaMA, Qwen, and Gemma.<n>We show that symbolic reasoning failures in LLMs are not due to missing representations or insufficient scale, but arise from structured interference within the model's computation graph.
  arXiv  Detail & Related papers  (2026-04-01T11:40:12Z)
• Reflection Pretraining Enables Token-Level Self-Correction in Biological Sequence Models [82.79223371188756]
  Chain-of-Thought (CoT) prompting has advanced task-solving capabilities in natural language processing with large language models.<n>Applying CoT to non-natural language domains, such as protein and RNA language models, is not yet possible.<n>We introduce pretraining, for the first time in a biological sequence model, which enables the model to engage in intermediate reasoning.
  arXiv  Detail & Related papers  (2025-12-24T05:25:17Z)
• Improving Procedural Skill Explanations via Constrained Generation: A Symbolic-LLM Hybrid Architecture [2.6141337419964725]
  In procedural skill learning, instructional explanations must convey not just steps, but the causal, goal-directed, and compositional logic behind them.<n>We present Ivy, an AI coaching system that delivers structured, multi-step explanations by combining symbolic Task-Method-Knowledge (TMK) models with a generative interpretation layer-an generative interpretation layer-that constructs explanations while being constrained by TMK structure.
  arXiv  Detail & Related papers  (2025-11-26T00:29:53Z)
• Temporal Sparse Autoencoders: Leveraging the Sequential Nature of Language for Interpretability [31.30541946703775]
  Translating internal representations and computations of models into concepts that humans can understand is a key goal of interpretability.<n>Recent dictionary learning methods such as Sparse Autoencoders provide a promising route to discover human-interpretable features.<n>But they exhibit a bias towards shallow, token-specific, or noisy features, such as "the phrase 'The' at the start of sentences"
  arXiv  Detail & Related papers  (2025-10-30T17:59:30Z)
• Schema for In-Context Learning [0.7850388075652649]
  In-context learning (ICL) enables language models to adapt to new tasks by conditioning on demonstration examples.<n>Inspired by cognitive science, we introduce SCHEMA ACTIVATED IN CONTEXT (SA-ICL)<n>This framework extracts the representation of the building blocks of cognition for the reasoning process instilled from prior examples.<n>We show that SA-ICL consistently boosts performance, up to 36.19 percent, when the single demonstration example is of high quality.
  arXiv  Detail & Related papers  (2025-10-14T21:00:15Z)
• LayerCake: Token-Aware Contrastive Decoding within Large Language Model Layers [53.43862310647276]
  Large language models (LLMs) excel at natural language understanding and generation but remain vulnerable to factual errors.<n>We introduce a token-aware, layer-localized contrastive decoding method that aligns specific token types with their most influential transformer layers to improve factual generation.<n>Our method requires no additional training or model modification, and experiments demonstrate that our method consistently improves factuality across multiple LLMs and various benchmarks.
  arXiv  Detail & Related papers  (2025-07-06T14:35:43Z)
• Sketch-of-Thought: Efficient LLM Reasoning with Adaptive Cognitive-Inspired Sketching [60.04718679054704]
  We introduce Sketch-of-Thought (SoT), a novel prompting framework.<n>It combines cognitive-inspired reasoning paradigms with linguistic constraints to minimize token usage.<n>SoT achieves token reductions of 76% with negligible accuracy impact.
  arXiv  Detail & Related papers  (2025-03-07T06:57:17Z)
• Causal Graphical Models for Vision-Language Compositional Understanding [36.24185263818946]
  We show that our method significantly outperforms all the state-of-the-art compositional approaches by a large margin.<n>It also improves over methods trained using much larger datasets.
  arXiv  Detail & Related papers  (2024-12-12T15:22:03Z)
• Chain-of-Thought in Large Language Models: Decoding, Projection, and Activation [8.80044898397965]
  Chain-of-Thought prompting has significantly enhanced the reasoning capabilities of large language models.<n>This work examines three key aspects: decoding, projection, and activation, aiming to elucidate the changes that occur within models when employing Chainof-Thought.
  arXiv  Detail & Related papers  (2024-12-05T07:47:29Z)
• Distributional Associations vs In-Context Reasoning: A Study of Feed-forward and Attention Layers [49.80959223722325]
  We study the distinction between feed-forward and attention layers in large language models.<n>We find that feed-forward layers tend to learn simple distributional associations such as bigrams, while attention layers focus on in-context reasoning.
  arXiv  Detail & Related papers  (2024-06-05T08:51:08Z)
• Explanation-aware Soft Ensemble Empowers Large Language Model In-context Learning [50.00090601424348]
  Large language models (LLMs) have shown remarkable capabilities in various natural language understanding tasks. We propose EASE, an Explanation-Aware Soft Ensemble framework to empower in-context learning with LLMs.
  arXiv  Detail & Related papers  (2023-11-13T06:13:38Z)
• Interpreting Pretrained Language Models via Concept Bottlenecks [55.47515772358389]
  Pretrained language models (PLMs) have made significant strides in various natural language processing tasks. The lack of interpretability due to their black-box'' nature poses challenges for responsible implementation. We propose a novel approach to interpreting PLMs by employing high-level, meaningful concepts that are easily understandable for humans.
  arXiv  Detail & Related papers  (2023-11-08T20:41:18Z)
• Simple Linguistic Inferences of Large Language Models (LLMs): Blind Spots and Blinds [59.71218039095155]
  We evaluate language understanding capacities on simple inference tasks that most humans find trivial. We target (i) grammatically-specified entailments, (ii) premises with evidential adverbs of uncertainty, and (iii) monotonicity entailments. The models exhibit moderate to low performance on these evaluation sets.
  arXiv  Detail & Related papers  (2023-05-24T06:41:09Z)
• Physics of Language Models: Part 1, Learning Hierarchical Language Structures [51.68385617116854]
  Transformer-based language models are effective but complex, and understanding their inner workings and reasoning mechanisms is a significant challenge.<n>We introduce a family of synthetic CFGs that produce hierarchical rules, capable of generating lengthy sentences.<n>We demonstrate that generative models like GPT can accurately learn and reason over CFG-defined hierarchies and generate sentences based on it.
  arXiv  Detail & Related papers  (2023-05-23T04:28:16Z)
• Autoregressive Structured Prediction with Language Models [73.11519625765301]
  We describe an approach to model structures as sequences of actions in an autoregressive manner with PLMs. Our approach achieves the new state-of-the-art on all the structured prediction tasks we looked at.
  arXiv  Detail & Related papers  (2022-10-26T13:27:26Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.