We already have…

• Maps of railroad lines, canals, etc.
• Over time

… but we can do more

Limited points in time

• Dependent on reliable line maps
• Rate of growth from 1870 and 1910
• Limits research designs

Lines vs. points

• Where does access to rail network happen?
• Do lines actually intersect?

… but we can do more

Owners and operators

• Politics and economics of largest companies in 19th century
• Who owns and operates these railroads?

What is it?

• Trade publication
  • frequently updated, for booking travel/freight
• Monthly/quarterly updates
• Years: 1868 through entire 20th century.

• (We start with 1880 to 1910, but more is possible)

What does it include?

• Hire RAs/firm to transcribe each page

Drawbacks

• Expensive, even more for multiple transcribers
• Cumbersome task: densely packed text
• Many errors requiring cleaning

• Train machine to read text
• Extract data by parsing text
• Automatic, cheap, fast

Drawbacks

• Even with custom training, performance is poor
  • small fonts, cheap printing
• Required extensive manual cleaning
• Automated Parsing is tricky

Making user interface is flexible/easy

• Combine images with a custom set of tasks (easier than MTurk/Google Form)
• Can customize many of the tasks

Discretize tasks:

• Divide labor across tasks/workers
• Less complexity \(\xrightarrow{}\) fewer errors, less training required, greater speed

Multiple workers:

• By default, multiple users classify/transcribe the same image
• Improved quality
• Volunteers can keep costs down

Community:

• Develop community of users committed to the project
• Interest in taking on more tasks related to the project
• Adding more ambitious goals over time

• Classification of items in an image
• Transcription of text
  • Limited data fields (forms)
  • Irregular format (telegraphs, letters, diaries)

Like many social scientists:

• Our data is tabular and dense
• Hundreds of items per page
• Same "fields" repeated

Task is overwhelming

• One page might take an hour or more

How can we …

1. Make transcription task simpler
2. But keep the work engaging for volunteers

1. Built custom tool for segmenting images

2. Interactive maps to make transcription work exciting

Automatic splitting?

• Splitting columns: Errors a problem
• Splitting rows: performs well

Procedure

1. Users identify page type
2. Users mark columns on the page
3. After \(n\) users classify
4. K-means clustering for column boundaries
5. Split page into columns
6. Automatic splitting of rows into columns

Key details

• Each resulting "cell" retains its page, its column index, and its row index
• "Cell" images automatically uploaded to Zooniverse
• Users see one row of data at a time

At work

Result of splitting

Multi-line station/telegraph entries

• How to aggregate data back to station/office?

Multi-line station/telegraph entries

Solution

Column indices, row indices, and row classification workflow

Lack of Context

What do these mean?

Lack of Context

Solution:

Tutorials/explanations

https://www.zooniverse.org/projects/mdweaver/wires-and-rails

Alienation from Labor

Volunteers do not follow a single station through

• split across rows
• different transcription tasks

Solution:

Interactive visualization:

• Map completed stations
• Searchable by user name
• Highlight/summarize contributions
• Ranking?

Larger projects

1. higher upfront cost
2. greater return when the scale is large

"Sparse" data

1. Easy to use when less data per image

"Dense" data

• Can build on our image segmentation
  • customization needed, but path is easier
• Requires planning to:
  • retain image metadata
  • eliminate redundancies in workflows
• Helpful with irregular formatting

Even when crowdsourcing seems too much…

• Workflow tools, discrete tasks, multiple users can be simplify work of paid RAs

Once transcriptions are completed, volunteers can:

• check quality of geocoding
• match company names and abbreviations

New data for the project:

• Time Tables (what kind of service, how frequent)

New data for the project:

• Time Tables (what kind of service, how frequent)
• Rail Postal Roads