Collection search and online publishing
Search that finds the exact word and the same idea in other words, shows the hit on the page image, and works on a site we host or inside the system you already run.
Digitized collections are often hard to search. A researcher types a surname into a portal, gets nothing back because the name is only in the handwriting, and leaves. Or the search finds the page but shows a thumbnail, and the reader has to hunt for the word. We build search that reads the recognized text and the catalog records, finds both exact words and related ideas, and opens the page image with the hit marked.
Two kinds of search, in plain words
Full-text search looks for the words you type. It is the right tool for names, places and exact phrases, and it is predictable: a result contains your word, or it does not appear. Its weakness is spelling. Old documents spell names three ways, and recognized text has errors, so we add tolerant matching for near misses and let you load known variant spellings.
Meaning-based search, also called semantic search, looks for passages about the same thing even when they use other words. A language model turns each passage, and each query, into a list of numbers that represents its meaning. Passages whose numbers sit close to the query’s are returned, whatever words they use. It is good for questions a researcher cannot phrase in the document’s own vocabulary. Its weakness is that it can return near neighbors that are not relevant, so its results are marked as matched by meaning, and exact matches stay at the top.
Example (fictional). In the Aster County Historical Society’s Box 3, nobody wrote the word “strike”. A full-text search for it finds nothing. A meaning-based search finds Ida Okerlund’s letter of February 3, 1894 (“The smelter men walked out on Monday”) and the Wren Gulch Courier notice that “the trouble at the smelter ended Thursday”. The search on the example collection page shows the idea with a small hand-made table of related words; a real index uses a language model instead.
Seeing the hit on the page
Search results open on the page image, not just the text. Our text recognition delivers ALTO XML, the Library of Congress layout standard, which records the position of every word on the page. The viewer uses those positions to highlight each hit on the image itself, so the reader sees the word in the original hand and can judge the reading. Text that has not been corrected is labeled as uncorrected, so nobody mistakes a machine reading for a transcription.
IIIF viewing
We publish images through IIIF, the International Image Interoperability Framework. The Image API 3.0 serves deep zoom, so a reader can open a whole map and zoom to one lot number without downloading a large file. The Presentation API 3.0 describes each item as a manifest: page order, labels and structure. Both are the current stable versions; Presentation API 4.0 is a release candidate, and we will follow it when it becomes stable. The IIIF Content Search API 2.0 lets a viewer search inside a single item.
Because these are open specifications, any IIIF viewer can open your manifests, and other institutions can show your items in their own online exhibitions while the images are still served from your site.
Two ways to publish
| Hosted search site | Import into your system | |
|---|---|---|
| What you get | A search site for your collection, run by us | Files prepared for the system you already use |
| Search | Full-text and meaning-based, with IIIF viewing | Whatever your system offers, plus our text and records |
| Best for | Collections with no public platform yet | Collections with an established catalog or portal |
| Formats | Built from your masters, text and records | CSV for Omeka’s CSV Import, EAD or MARCXML for ArchivesSpace, IIIF manifests where supported |
Hosting is optional. The cataloging behind both routes is the same, and it is described under AI-assisted cataloging.
Leaving a platform with all files
A hosted site should never hold your collection hostage. If you stop hosting with us, you leave with the preservation masters, access images, text in ALTO and plain text, catalog records in CSV and standard XML with their review labels, IIIF manifests, and a description of how the search index was configured. The files are packaged with checksums as part of digital preservation, so you can verify that nothing was lost in the move. Hosting terms, and ownership and licensing of the files, are set out in each agreement.
Accessibility of the viewer
We build search pages and viewers to WCAG 2.1 Level AA and test them with a keyboard and a screen reader before launch. Every control can be reached without a mouse, images have text alternatives, and the transcription sits next to the page image as real text, so a screen reader user can read a letter that sighted readers see in handwriting.
For state and local governments this is also a legal question. The Justice Department’s ADA Title II rule requires their web content to meet WCAG 2.1 AA, with compliance dates now set at April 26, 2027 for public entities with a population of 50,000 or more, and April 26, 2028 for smaller entities and special districts. The rule has a narrow exception for archived web content. Our page on local government records explains what that means for digitized records, and your counsel decides how it applies to you.
To see search working on real-looking records, try the example collection, or read the Reading Room’s text recognition and search section.
Send one box first.
A pilot box shows you the images, the recognized text and the catalog records before you commit a whole collection. Tell us what is in the box and what you want people to find.
Studio and contacts
24 South Weber Street, Suite 225Colorado Springs, CO 80903(719) 244-8450sales@dastok.com