arXiv：cs.CL· Sean Diab·· 6 小时前
Reviser：用自回归光标动作（插入、移动、删除、停止）写作与编辑文本
Reviser: Writing and Editing Text with Autoregressive Cursor Actions
AI 导读
研究者提出 Reviser，一个 decoder-only Transformer，通过光标动作（插入、移动、删除、停止）在画布上写作和编辑文本，由确定性执行器执行，且对自身动作自回归，单步开销与 AR 步骤相同。
正文
Abstract:Language models write text once, left to right, and models that write in other orders, such as diffusion LMs, pay for it with many full-sequence passes. We introduce Reviser, a decoder-only Transformer that writes and edits text with cursor actions (insert, move, delete, stop) that a deterministic executor applies to a canvas. Reviser is autoregressive over its own actions, so a step costs the same as an autoregressive (AR) step. We train Reviser to continue text, including a self-correction recipe that types a wrong token and deletes it, and Reviser-Edit to revise a draft, and compare each with a same-size AR model trained on the same data, using three LLM judges on continuations cut to the same length. Reviser-Edit beats its matched AR rewriter on CoEdIT SARI (+7.2 at 100M, +4.5 at 300M; two seeds agree) with a third of the decoding steps and latency, though the AR rewriter corrects grammar better. In continuation, Reviser beats diffusion LMs of similar size (62-74% win rate) at a fraction of their compute; two of three judges prefer its AR twin, and one slightly prefers Reviser at 100M. Reviser fills gaps it left and backspaces a just-typed repeated token with probability 0.85-0.87.
| Comments: | 42 pages. Code: this https URL. Checkpoints: this https URL. Trajectory replays: this https URL |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2609.20830 [cs.CL] |
| (or arXiv:2609.20830v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2609.20830 arXiv-issued DOI via DataCite |
Submission history
From: Sean Diab [view email]
[v1]
Thu, 23 Jul 2026 00:54:26 UTC (82 KB)
[v2]
Wed, 14 Oct 2026 03:50:37 UTC (109 KB)
来源：arXiv：cs.CL · arxiv.org