Why People Actually Use Digitized Obituaries
I keep running into the same question from folks who stumble onto newspaper archives looking for family connections. The answer isn't as straightforward as you might expect. Obituaries aren't just names and dates. They're fragments of what a neighborhood actually looked like, who lived there, who visited who, and what churches or union halls mattered at the time. The problem is most of this stuff sat in physical binders in library basements for decades, deteriorating quietly. Digitization changed that, but it didn't make things easy by any means. The New Jersey Herald ran obituaries for over a century, and when those pages were scanned and indexed, they became something researchers can actually search instead of squinting at microfiche until their eyes give out. The project itself was a collaboration between the newspaper publisher, a county historical society, and a digitization vendor who specialized in newspaper archives. The technical side involved spectral imaging to pull text from papers that had yellowed or faded, then running everything through OCR with custom dictionaries tuned to 19th and early 20th century naming conventions. That last part matters more than people realize. Standard OCR keeps failing on names like "Schneiderhan" or "McGarrity" because the training data doesn't account for regional spelling variations. The custom dictionary fixed maybe sixty percent of those failures, which sounds bad until you remember that sixty percent represents thousands of previously unrecoverable records. Here's how the preservation process actually works in practice. First, the physical volumes are boxed and moved to a climate-controlled staging area. Each page gets photographed with a high-resolution overhead scanner calibrated for color fidelity on aged paper. The images go through a preprocessing step where desaturation and contrast stretching happen before OCR even touches them. Then the text layer gets attached to the image in a searchable database format, usually something like CDWA or a custom XML schema. The indexing part is where the real work shows. Name variants get cross-referenced, addresses get geocoded where possible, and familial relationships mentioned in the text get flagged for later analysis. This takes significantly longer than the scanning itself, usually three to four hours of manual review per hundred obituaries depending on legibility.
I ran into a specific edge case last year that I haven't seen documented anywhere. There was a batch of obituaries from the 1898 flu season where the typesetter had used abbreviated middle initials inconsistently across the same article. One obit would list "John W. Miller" and the next would have "Johnson W. Miller" for the same person because the family submitted two separate copies to different editors. Standard deduplication algorithms would treat these as different people. What I ended up doing was writing a quick Python script that matched on address plus approximate date of death plus surviving family names, then flagged those clusters for manual verification. Cut the dedup time from roughly twenty minutes per record to about ninety seconds per cluster. Not elegant, but it worked for the forty-seven duplicate entries I found in that batch alone. The metadata structure is worth paying attention to if you're planning to do anything beyond casual browsing. Each obituary record typically contains the full text, scan timestamp, source volume number, page number, OCR confidence score, and a set of extracted entities including person names, locations, organizations, and dates. The confidence score is the part most people ignore. An OCR confidence below seventy percent usually means the text has errors that will throw off name searches. I always filter for confidence scores above that threshold when I'm doing serious research, then manually check the image for anything that falls below it. Takes a bit longer but prevents chasing ghosts through bad text. There are real limitations to this kind of digitization project that the press releases never mention. The biggest one is coverage gaps. The Herald didn't publish every week consistently before 1923, and there are entire months missing from certain years where the physical copies were lost to water damage or simply never archived. No amount of OCR fixes that. Second, the OCR quality degrades noticeably on anything printed on cheap newsprint from the 1940s onward, where the ink absorption created blurred characters that even custom dictionaries can't rescue. You'll find yourself going back to the original images constantly, which defeats some of the purpose of having a searchable database.
If you're approaching this as a researcher, the most useful thing you can do is start with known names and work outward. Pick an ancestor, find their obit, then follow the surviving family members mentioned in it to nearby obituaries. That chain often reveals connections that keyword searches miss entirely. The database doesn't link records automatically, so you're doing the graph traversal yourself. It's slower but more accurate than relying on search suggestions, which tend to surface tangentially related results that aren't actually connected. A couple of counter-intuitive things I've learned from working with these archives. First, the most valuable obituaries aren't always the longest ones. A brief notice from 1902 might mention a person's birthplace, their occupation, and three relatives who survived them. That's enough to place someone in a specific township and connect them to a broader family network. Second, the index terms the database uses sometimes differ from what appears in the actual text. The catalogers added subject headings based on their own interpretation, which can lead to mismatches. Always verify by opening the full scan rather than trusting the filtered results at face value. I've wasted hours tracking down people who weren't actually in the records I thought I'd found because the subject heading was wrong. The project itself is ongoing, with new batches being added periodically as physical copies are located and processed. There's no fixed completion date because the source material isn't entirely accounted for yet. Some volumes are still held by private collectors or local genealogical societies who haven't signed agreements to contribute their holdings. If you're interested in the archive, the best approach is to contact the managing institution directly and ask about both the current access points and any restricted collections that haven't been digitized yet. The answer you get will tell you more about what's actually available than any published finding aid.
Get the Full Details
