El problema
Why immigration manifest OCR fails on ship passenger lists
Immigration records are among the most challenging documents in genealogy. Ship manifests were filled out by port inspectors and ship clerks writing under pressure, often recording names from dozens of languages they didn't speak. A single Ellis Island manifest page might contain Irish, Italian, Polish, Yiddish, and German names — each phonetically approximated by an English-speaking clerk. Earlier records from Castle Garden (1820-1892) are even harder: cramped columns, faded ink, and minimal standardisation. Standard OCR tools fail because they can't handle the mix of languages, the dense tabular layout, and the heavily abbreviated entries.
Multi-language names on every page — clerks wrote what they heard, not what was spelled
Dense tabular layouts with 20+ columns on a single manifest sheet
Heavy abbreviation: 'do.' for ditto, country codes, occupation shorthand, tick marks
Faded ink, pencil annotations, stamps, and microfilm artefacts obscure entries
Names transliterated from Cyrillic, Hebrew, Arabic, and other scripts into Latin letters
La solución
AI that reads Ellis Island record handwriting and passenger arrival records
Transkribus uses handwritten text recognition (HTR) trained on millions of historical handwriting samples — including immigration-era documents. Unlike standard OCR, it understands connected cursive, dense tabular layouts, and the abbreviation conventions used on ship manifests and naturalization forms. The AI handles 100+ languages and scripts, making it uniquely suited for the multilingual reality of immigration records. Upload a scan or photo, select a model, and get structured, searchable text back in minutes.
300+ public models — including models for 19th and 20th-century handwriting styles used on manifests
100+ languages and scripts — reads names originally written in Latin, Cyrillic, Hebrew, and more
Table detection identifies manifest columns: name, age, occupation, nationality, last residence, destination
Confidence scores on every entry — see exactly which names and fields need your review
Export as plain text, CSV, Excel, searchable PDF, or structured XML
How to transcribe immigration records in 4 steps
Subir su documento
Take a photo with your phone or upload a scan. Transkribus accepts JPG, PNG, PDF, and TIFF — including multi-page manifests and naturalization packets.
Seleccionar un modelo de IA
Choose from 300+ public models. Filter by language, century, and script type to find the best match for your immigration record's era and origin.
Ejecutar reconocimiento de texto
Click 'Recognise' and let the AI work. It detects the manifest's table structure, reads each cell, and processes the full page in about 30 seconds.
Revisar y exportar
Check the transcription against the original image. Fix any misread names in the built-in editor, then export as text, CSV, Excel, PDF, or XML.
Models for immigration records
AI models for ship passenger list transcription and naturalization records
The Transkribus public model catalog includes models suited to the handwriting styles found on immigration documents from major ports and eras. These models understand the dense columnar layouts, standardised forms, and multilingual entries that make immigration records unique among genealogical sources.
US port records: models for Ellis Island-era manifest handwriting (1892-1954) and earlier Castle Garden records
UK and European port records: models for Liverpool, Hamburg, Bremen, and Le Havre passenger lists
Naturalization papers: models for handwritten petitions, declarations of intention, and certificates
Multi-language support: 100+ languages, critical for records mixing English, German, Italian, Polish, Yiddish, and more
New models added by the community regularly — covering more ports, eras, and document types
Preguntas frecuentes
Ready to unlock your immigration records?
Create a free account and start transcribing ship manifests and passenger lists today. 50 free credits every month — no credit card required.
50 créditos gratuitos al mes — Sin tarjeta de crédito
200 M+Páginas procesadas
500.000+Usuarios en todo el mundo
300+Modelos de IA públicos