Use of artificial intelligence (AI) in historical records transcription: Opportunities, challenges, and future directions
Bibliographic record
Abstract
Historical weather documents are hidden treasures for understanding long-term climatic changes that were not fully exploited in the past due to limited resources.They are now attracting much attention from data rescue researchers who are interested in the valuable information the historical documents can bring.One obstacle of researching these historical documents is that the documents can only exist in hard copy format, making it challenging to access computational and geographically.The most common transcription approach is manual transcription.Manual transcription is labour-intensive and can be time-consuming when transcribing large numbers of documents, especially when they are written in cursive handwriting and in dense ledge sheets.If the transcription could be automated, considerable time and resources would be saved.The overarching question driving this research is what is the role and benefit of automation, especially artificial intelligence (AI)-augmented automation, in historical weather data rescue?To answer this question, I divide it into two complementary questions: process raw tabular images, segment tabular cells, recognize the data entries, and rearrange the results into original formats.The proposed workflow is tested and evaluated on the Data Rescue: Archives and Weather (DRAW) dataset.The proposed workflow provides a guideline for researchers who want an end-to-end solution on automatic historical data rescue and it hopefully can be replicable for future projects that seek automated data rescue.This thesis makes several contributions to the use of AI in historical weather data rescue and historical records transcription more generally.First, this study contributes by uncovering the opportunities and challenges of using AI in historical records transcription.Second, this research contributes a benchmark workflow for AI-augmented historical records transcription that can be customized and adapted to different types of historical records.Third, this study identifies and bridges the gap between the use of AI and the rescue and transcription of historical records, finding that addressing this multi-disciplinary problem requires efforts from different fields and communities.The results of this study will serve as a starting point for future studies who want to involve AI in their transcription process and as a reference for future attempts.iii Ré sumé Les documents mé té orologiques historiques sont des tré sors caché s pour la compré hension des changements climatiques qui n'ont pas é té pleinement exploité s dans le passé en raison de ressources limité es.Ils attirent aujourd'hui l'attention des chercheurs qui s'inté ressent aux informations pré cieuses que les documents historiques peuvent apporter.L'un des obstacles à la recherche de ces documents historiques est qu'ils n'existent que sous forme de copie papier, ce qui en rend l'accè s difficile.La mé thode de transcription la plus courante est la transcription manuelle.La transcription manuelle demande beaucoup de travail et peut prendre beaucoup de temps lorsqu'il s'agit de transcrire un grand nombre de documents.Si la transcription pouvait ê tre automatisé e, cela permettrait d'é conomiser beaucoup de temps et de ressources.La question primordiale qui sous-tend cette recherche est la suivante : quel est le rôle et les avantages de l'automatisation, en particulier de l'automatisation augmenté e par l'intelligence artificielle (IA), dans le sauvetage des donné es mé té orologiques historiques ?Pour ré pondre à cette question, je la divise en deux questions complé mentaires : 1. Comment les chercheurs et les praticiens perç oivent-ils les dé fis et les opportunité s de l'utilisation de l'IA dans le sauvetage des donné es ? 2. Si le sauvetage des donné es par l'IA est utile, à quoi pourrait ressembler un systè me automatisé ?Je commence par passer en revue la litté rature sur le sauvetage des donné es mé té orologiques historiques et, plus gé né ralement, sur la transcription des documents historiques.Il n'existe pas de flux de travail systé matique pour guider les chercheurs dans l'automatisation du processus de transcription.Il existe é galement des risques et des dé fis ; la faç on dont les chercheurs les perç oivent est inconnue.Au chapitre 3, j'ai mené une enquê te sur l'attitude et l'opinion des chercheurs en sauvetage de donné es et en science citoyenne à l'é gard de l'utilisation d'approches amé lioré es par l'IA plutôt que d'approches manuelles.Le ré sultat fournira une perspective de l'adaptation de l'IA dans le sauvetage des donné es.Les ré pondants ont suggé ré qu'un modè le hybride où l'homme et l'IA travaillent ensemble serait pré fé rable à une approche purement IA.Au chapitre 4, j'ai cré é et testé un flux de travail de sauvetage de donné es enrichi par l'IA, qui peut automatiser la transcription d'un ensemble de donné es tabulaires manuscrites historiques.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.037 | 0.083 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.003 | 0.005 |
| Science and technology studies | 0.001 | 0.003 |
| Scholarly communication | 0.013 | 0.012 |
| Open science | 0.002 | 0.003 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.006 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".