Enterprise AI Analysis
A multi-modal dataset and method for bone-level association prediction in oracle bone inscriptions
This research introduces the first public benchmark dataset and a novel multi-modal deep learning method for predicting bone-level associations in oracle bone inscription sentences. It aims to address the challenges of fragmentation and incomplete contextual information, significantly enhancing the digital reconstruction and understanding of ancient Chinese texts.
Executive Impact
By providing accurate bone-level association predictions, this AI method dramatically improves the efficiency and reliability of oracle bone rejoining, unlocking historical insights previously fragmented. This innovation offers significant value for archaeological, linguistic, and historical research, accelerating the understanding of ancient Chinese civilization.
0.9587
AUROC Score
0.7390
F1 Score
9,935 Inscriptions
Dataset Size
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
The Oracle Bone Inscription Dataset with Additional Contextual Reconstruction (OBID-ACR) is the first public benchmark for bone-level association prediction. It integrates glyph images, OBI sentences, and primary/secondary character tags, addressing limitations of previous datasets by focusing on multi-modal information for fragmented oracle bones.
3551
Oracle Bone Fragments in OBID-ACR
Dataset Construction Process
Extract OBI Sentences from OBIMD
→
Retain Fragments with ≥2 Characters
→
Identify Additional Rejoined Cases
→
Compile Final OBID-ACR Dataset
The proposed Siamese BiLSTM with Glyph-based Embeddings for Bone-level Sentence Association Prediction (SGBSAP) uses a VAE to learn character-level representations from glyph images. A Siamese dual-tower BiLSTM network then processes sentence pairs for association prediction, outperforming context-based methods.
0.9587
SGBSAP AUROC (Glyph-based)
| Strategy | Benefits | Limitations |
|---|---|---|
| Glyph-based Embeddings (VAE) |
|
|
| Context-based Embeddings (SGNS/CBOW) |
|
|
| Combined Multi-modal (SGBSAP-Weighted) |
|
|
Case studies demonstrate SGBSAP's effectiveness in rejoining fragmented oracle bones, even when physically distant. It correctly identifies same-bone relationships and provides high association scores, though it shows limitations with substantial character loss.
Rejoining Directly Connected Fragments (H51 & H64)
Context: Fragments H51 and H64, from the Oracle Bone Inscription Collection, were directly connected. Sentence (a) from H51: 'Divination: do many people perish in a certain locality because of warfare?'; Sentence (b) from H64: 'Divination: heavy personnel losses in warfare, prayers for divine protection.' These sentences clearly address the same underlying issue, and the broken edges and handwriting style were consistent.
Findings: SGBSAP achieved an association score of 0.9996, ranking within the top 1.49% of all sentence pairs in the test dataset. This result strongly indicates these fragments can be rejoined.
Rejoining Not Connected Fragments (H30107 & H30109)
Context: Fragments H30107 and H30109, from the Oracle Bone Inscription Collection, were not directly connected but belong to the same source bone. Sentence (a) from H30109 and sentence (b) from H30107 are basically identical in structure and theme: 'Divination: the king of Shang should not perform rain sacrifice in July.' Character glyph styles were very consistent.
Findings: SGBSAP achieved an association score of 0.9550, ranking within the top 5.81% of all sentence pairs. This indicates these fragments can be rejoined.
Limitation Example (H16756 & H16773)
Context: Fragments H16756 and H16773, from the Oracle Bone Inscription Collection, were rejoined by experts. Sentence (a) from H16756 records a divination concerning whether disasters occur within the next ten days. Sentence (b) from H16773 exhibits a similar syntactic pattern but with substantial missing characters. The missing content leads to significant loss of contextual information.
Findings: SGBSAP achieved an association score of 0.0102, ranking within the top 12.72% of all sentence pairs. This indicates that SGBSAP incorrectly assessed their association, highlighting limitations with substantial character loss and reliance on contextual completeness.
Calculate Your Potential ROI
Estimate the efficiency gains and cost savings your enterprise could realize by implementing AI-driven solutions.
Estimated Annual Savings
$0
Annual Hours Reclaimed
0
Implementation Roadmap
A structured approach to integrating cutting-edge AI into your enterprise workflows.
Phase 1: Data Acquisition & Preprocessing
Compile and clean the multi-modal Oracle Bone Inscription Dataset (OBID-ACR), ensuring authenticity and quality.
Phase 2: Glyph Embedding Model Development
Design and train the Variational Autoencoder (VAE) to learn robust character-level glyph embeddings from images.
Phase 3: Association Prediction Model Training
Train the Siamese BiLSTM network using glyph-based embeddings to predict bone-level associations between sentence pairs.
Phase 4: Empirical Evaluation & Case Studies
Conduct extensive experiments, compare with baselines, and validate the model's performance on real-world oracle bone rejoining cases.
Phase 5: Deployment & Integration
Integrate the SGBSAP model into archaeological and historical research tools for digital reconstruction and interpretation.
Ready to Transform Your Enterprise?
Schedule a personalized consultation to explore how our AI solutions can be tailored to your specific needs.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat