Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note2 min read

Multi-Expert Routing for Multi-Domain Low-Resource OCR: A Manchu Case Study

Source

arXiv

arxiv.org/abs/2607.14041v1

Read the source

Source

Plain-Language Brief

The paper introduces a multi-expert system for Optical Character Recognition (OCR) of historical Manchu texts which have various writing styles and limited labeled data. The system uses a router that classifies pages by visual style and assigns them to suitable domain specialists. The results show that the routed system accurately matches the selected specialist for each style in three tests, with high page-level domain accuracy and matching the domain-label oracle at similar precision.

What the Document Says

  • The paper focuses on historical Manchu OCR.
  • The texts have various visually distinct writing styles: regular script, running script, and chancery hand (used in palace memorials).
  • The authors propose a multi-expert system that reuses checkpoints from an iterative fine-tuning process as domain specialists.
  • A lightweight page-level image classifier is used to dispatch pages by visual style.
  • When no suitable specialist exists, the system trains an additional expert for that domain.
  • On three test sets, the routed system matches the selected specialist for each style with low Character Error Rate (CER): 0.30 percent on regular script, 1.57 percent on memorials, and 4.83 percent on running script.
  • The router achieves 99.3 percent page-level domain accuracy and matches the domain-label oracle at similar precision.
  • Two of the three selected specialists were not specifically trained for their final domain; only the running-script expert was.

Key Concepts Mentioned

  • historical Manchu OCR
  • multi-expert system
  • visually distinct writing styles
  • iterative fine-tuning
  • lightweight page-level image classifier
  • Character Error Rate (CER)
  • page-level domain accuracy

Claims Made by the Document

  • The proposed multi-expert system can accurately match the selected specialist for each style in historical Manchu OCR.
  • The router achieves high page-level domain accuracy.

Evidence Used

  • Results on three test sets

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Historical Manchu OCR must accommodate various visually distinct writing styles, including regular script, running script, and the semi-cursive chancery hand used in palace memorials, despite limited labeled data. We study a multi-expert system that reuses checkpoints from an iterative fine-tuning process as domain specialists and uses a lightweight page-level image classifier to dispatch pages by visual style. When the checkpoint pool lacks a suitable specialist, we train an additional expert for that domain. On three frozen test sets, the routed system matches the selected specialist for each style at two-decimal precision: 0.30 percent CER on regular script, 1.57 percent on memorials, and 4.83 percent on running script. The router achieves 99.3 percent page-level domain accuracy and matches the domain-label oracle at the same precision. Two of the three selected specialists were not trained specifically for their final domain; only the running-script expert was trained with that domain as its target. We report the evaluation protocol, router design, and per-page predictions to make the comparison reproducible.