Zugwise: Domain-Aware Game Reconstruction for Digitizing Handwritten Chess Scoresheets

The digitisation of hand‑written chess games has traditionally been treated as a handwriting‑recognition problem, limited to identifying the move written in each square and verifying its legality. This approach fails to catch the most subtle mistakes: when the recogniser produces a legal but incorrect move, the process continues without flagging the error, leading to a wrong reconstruction of the rest of the game.

To overcome this limitation, the authors introduce Zugwise, a system that reframes the task as a full‑game reconstruction. It leverages chess‑specific algorithms—such as quiet‑search, static exchange evaluation, and pin detection—to ask whether the sequence written by the player is plausible. An absurdity‑detection module flags “silent” errors as soon as they appear, and a retrospective search tracks the root cause of later‑detected failures.

The method merges the second player’s sheet using a Needleman‑Wunsch‑style alignment and forwards only the discrepancies between the two legal readings to human review. In trials on two tournament sections, incorporating these components reduced silent errors from 16 to 1 and from 50 to 13, respectively. Moreover, applying two learned signals—a CTC re‑scoring on the recogniser output and a human‑move probability model similar to Maia—raised the correct‑correction rate at the highest position from 58.8 % to 78.0 % in cases where an alternative was viable.

This progress is significant because it enhances the reliability of automatic game transcription, a critical step for archiving, analysing, and sharing historic games without relying on a single source of truth. The solution runs entirely in the browser, is free, respects user privacy, and works offline, making it accessible to players, coaches, and researchers.

Read the original article on arXiv cs.AI

This summary is an informational synthesis produced by dataqbs.com. All rights to the original content belong to its author and the cited media outlet. We act solely as curators of technology news and claim no authorship.