SPOT: Span-level Pause-of-Thought for Efficient and Interpretable Latent Reasoning in Large Language Models

• URL: http://arxiv.org/abs/2603.06222v1
• Date: Fri, 06 Mar 2026 12:34:27 GMT
• Title: SPOT: Span-level Pause-of-Thought for Efficient and Interpretable Latent Reasoning in Large Language Models
• Abstract summary: Explicit Chain-of-Thought (CoT) computation incurs high inference cost due to internalize token-level traces.<n>We propose SPOT, a flexible framework that compresses explicit CoT into compact latent pause tokens without enforcing a fixed response template.<n>Experiments on reasoning benchmarks demonstrate that SPOT improves accuracy by 2.3 points on average while reducing generated tokens by 37.5%.
• Score: 15.95627037350657
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Explicit Chain-of-Thought improves the reasoning performance of large language models but often incurs high inference cost due to verbose token-level traces. While recent approaches reduce this overhead via concise prompting or step pruning, they largely truncate what the model says rather than internalize what the model thinks. Latent reasoning offers a promising alternative by performing computation in the hidden space, yet prior methods face two critical challenges. Many existing approaches rely on rigid point-to-point alignment, forcing a latent token to approximate the final representation of a reasoning step, which can be insufficient to capture the dense, variable-length semantics of an entire reasoning segment. Furthermore, these methods often suffer from a lack of interpretability: latent states are commonly produced by unconstrained optimization or embedding mixing, yielding vectors that are difficult to decode or audit under the pretrained language head. We propose SPOT, a flexible framework that compresses explicit CoT into compact latent pause tokens without enforcing a fixed response template. At the core of SPOT is Span-level Semantic Alignment, a Sinkhorn optimal-transport objective that softly matches each pause token to the semantics of an entire reasoning segment, overcoming the rigidity of step-end alignment. To further improve interpretability, SPOT introduces a Frozen-Head Decoding Constraint that keeps latent states directly decodable as token distributions under the frozen pretrained LM head, enabling readable keyword interpretations of latent thoughts. Experiments on reasoning benchmarks demonstrate that SPOT improves accuracy by 2.3 points on average while reducing generated tokens by 37.5% and provides faithful semantic interpretations of the latent reasoning process.

Related papers

• Latent Recurrent Thoughts: Recurrent Refinement of Proposed Latents for Reasoning with Frozen LLMs [16.72007494160683]
  Chain-of-thought reasoning unfolds in discrete token space.<n> Reasoning instead in a model's continuous representation space.<n>We instantiate this as Latent Recurrent Thoughts.
  arXiv  Detail & Related papers  (2026-09-01T11:56:17Z)
• CoLT: Teaching Multi-Modal Models to Think with Chain of Latent Thoughts [25.917787556467776]
  Chain-of-thought (CoT) reasoning has enabled multi-modal large language models (MLLMs) to tackle complex visual reasoning tasks.<n>We propose CoLT, a novel framework that teaches multi-modal models to reason through a chain of latent thought representations.<n>Compared to text CoT methods, CoLT can notably reduce the inference time by 10.1$times$ and text decoding time by 22.6$times$.
  arXiv  Detail & Related papers  (2026-06-30T17:24:40Z)
• Why Struggle with Continuous Latents? Interpretable Discrete Latent Reasoning via Rendered Compression [42.15219720606396]
  Large language models achieve high reasoning performance via explicit chain-of-thought and reinforcement learning.<n>Latent reasoning reduces this cost by shifting computation into a latent space.<n>We propose textbfDiscrete Latent Reasoning(DLR), the first method that converts continuous latent states into explicit discrete tokens.
  arXiv  Detail & Related papers  (2026-06-29T02:34:52Z)
• Latent Thoughts Tuning: Bridging Context and Reasoning with Fused Information in Latent Tokens [13.653741247835091]
  Latent Thoughts Tuning (LT-Tuning) is a framework that redefines how latent thoughts are constructed and deployed.<n>We introduce a Context-Prediction-Fusion mechanism that jointly leveraging contextual hidden states and predictive semantic guidance.<n>Our method outperforms existing latent reasoning baselines, effectively mitigating feature collapse and achieving robust reasoning accuracy.
  arXiv  Detail & Related papers  (2026-02-10T19:19:10Z)
• Latent Chain-of-Thought as Planning: Decoupling Reasoning from Verbalization [9.193078163792427]
  Chain-of-Thought (CoT) empowers Large Language Models (LLMs) to tackle complex problems.<n>Recent latent reasoning approaches attempt to optimize efficiency by performing reasoning within continuous hidden states.<n>We introduce PLaT, a framework that reformulates latent reasoning as planning by fundamentally decouple reasoning from verbalization.
  arXiv  Detail & Related papers  (2026-01-29T07:38:18Z)
• Latent Reasoning in LLMs as a Vocabulary-Space Superposition [80.01651003144282]
  Large language models (LLMs) demonstrate strong reasoning abilities with chain-of-thought prompting, but explicit reasoning introduces substantial computational overhead.<n>Recent work on latent reasoning reduces this cost by reasoning in latent space without explicit supervision, but performance drops significantly.<n>To address this, we restrict the latent space to the column space of the LLM vocabulary, treating latent reasoning as a superposition over vocabulary probabilities.<n>Once latent reasoning concludes, it collapses into an eigenstate of explicit reasoning to yield the final answer.<n>Latent-SFT sets a new state of the art on GSM8k, matching explicit
  arXiv  Detail & Related papers  (2025-10-17T10:51:20Z)
• Stop Spinning Wheels: Mitigating LLM Overthinking via Mining Patterns for Early Reasoning Exit [114.83867400179354]
  Overthinking can degrade overall performance of large language models.<n>We categorize reasoning into three stages: insufficient exploration stage, compensatory reasoning stage, and reasoning convergence stage.<n>We develop a lightweight thresholding strategy based on rules to improve reasoning accuracy.
  arXiv  Detail & Related papers  (2025-08-25T03:17:17Z)
• ConciseHint: Boosting Efficient Reasoning via Continuous Concise Hints during Generation [74.37307916314407]
  We propose a framework dubbed ConciseHint, which continuously encourages the reasoning model to speak concisely.<n>Experiments on the state-of-the-art LRMs, including DeepSeek-R1 and Qwen-3 series, demonstrate that our method can effectively produce concise reasoning.
  arXiv  Detail & Related papers  (2025-06-23T16:20:44Z)
• Soft Thinking: Unlocking the Reasoning Potential of LLMs in Continuous Concept Space [62.54887038032942]
  We introduce Soft Thinking, a training-free method that emulates human-like "soft" reasoning by generating soft, abstract concept tokens.<n>These concept tokens are created by the probability-weighted mixture of token embeddings, which form the continuous concept space.<n>In essence, each generated concept token encapsulates multiple meanings from related discrete tokens, implicitly exploring various reasoning paths to converge.
  arXiv  Detail & Related papers  (2025-05-21T17:29:15Z)
• Fractured Chain-of-Thought Reasoning [61.647243580650446]
  We introduce Fractured Sampling, a unified inference-time strategy that interpolates between full CoT and solution-only sampling.<n>We show that Fractured Sampling consistently achieves superior accuracy-cost trade-offs, yielding steep log-linear scaling gains in Pass@k versus token budget.
  arXiv  Detail & Related papers  (2025-05-19T11:30:41Z)
• Sketch-of-Thought: Efficient LLM Reasoning with Adaptive Cognitive-Inspired Sketching [64.74765550805024]
  Chain-of-Thought prompting elicits step-by-step problem solving, but often at the cost of excessive verbosity in intermediate outputs.<n>We propose Sketch-of-Thought (SoT), a prompting framework that integrates cognitively inspired reasoning paradigms with linguistic constraints.<n>SoT achieves token reductions of up to 84% with minimal accuracy loss across 18 reasoning datasets.
  arXiv  Detail & Related papers  (2025-03-07T06:57:17Z)

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.