Search a Scanned PDF for a Name Without Missing Misreads

To search a scanned PDF for a name without missing pages, search for more than the name. Build a list of its spellings, the forms it takes on legal pages and the ways recognition is likely to misread it, run every entry, then search for the address, title or account number that travels with it.

A single search for "Brennecke" answers one narrow question: on which pages did the recognized text come out as exactly those nine letters? The pages where it came out as "Brermecke" stay silent. A name search you can defend is a set of searches, each one logged.

Take the names from the caption and from the pages where they repeat. Search the plaintiff, the defendant, and defense counsel, with a short form and a likely misreading of each. Open a hit and compare it with the page image. Run an expert witness named on a deposition cover through the same steps.

Why are names the words a scanned PDF search misses most?

Names are exposed to misses for two reasons you can see in your own file. A name is in no word list, so a reader that leans on known words has nothing to pull a doubtful letter toward. And names carry letter pairs and capitals that look like other characters once a page has been copied, faxed or scanned.

No word list to lean on

Some recognition software weighs what the shapes look like against which words exist. That helps with "agreement" and does nothing for "Brennecke", and it can push a surname toward an ordinary word that resembles it. The post on whether OCR uses a dictionary explains that mechanism; here it is enough to know that a name gets no such help.

Shapes that pass for other shapes

A search matches characters, not appearances. Two letters that touch on a worn page can be read as one different letter, and the text layer then holds a word you would never think to type.

Treat the table as candidates to try, not as a forecast. Which of them your file contains depends on the scan, the typeface and the software that read it, and the only way to find out is to search.

Names sit where pages are worst

Look at where a name appears on a legal page: under a signature, inside a caption box, across a stamp, in a fax header, in the small print of a service list. If those are the rough parts of your scan, the name is recognized less cleanly than the body text around it. Zoom in on three such spots in your own file before you decide how wide the variant list needs to be.

Building a list of spellings and likely misreadings for one name

Build the list in three layers: how the name is properly spelled, how people have written it in this matter, and how recognition could have stored it. Keep the list in a spreadsheet with a column for hits, so the search itself becomes a record.

A worked example

Suppose you are looking for Ilse Brennecke and her company, Brennecke & Rill Haulage LLC, in a 1,400-page scanned production.

Layer one: real spellings. Collect them from documents, not from memory. Pleadings, the engagement letter, a driver's licence and the company's registration may each spell the name differently.

• Brennecke, Brenneke, Brenecke, Brennicke
• Ilse, Ilsa, Else
• Rill, Rille

Layer two: forms used in the file. Read ten pages where you know the name appears and note what you see.

• I. Brennecke; Brennecke, Ilse; Ms. Brennecke
• B&R Haulage; Brennecke and Rill; BRH
• An email handle such as ibrennecke

Layer three: likely misreadings. Apply the character swaps to the hard parts of the name. "Brennecke" has a double n, two letters e and a k, each a place where a shape can go wrong.

• Brermecke, Brenneckc, Brcnnecke, Bremecke
• 8rennecke, Brennec ke, Bren necke
• RiII, Ri11, Rili

Then harvest. Open the pages where you can see the name with your own eyes, select it and paste it into a text editor. Whatever the paste gives you is what the text layer holds on that kind of page, and it belongs on the list more than any guess does. The post on seeing the text layer in a PDF shows ways to read that layer directly.

Search for the sturdy fragment

A fragment that survives the usual damage finds variants you did not list. For "Brennecke", "ecke" and "Brenn" each sit clear of one of the trouble spots. Search for a fragment with whole-word matching off, accept the noise, and read the hits. Add each new variant to the spreadsheet.

The notes column earns its place. A cluster of misreadings in one page range tells you that batch was scanned badly, and that every other search you run over it deserves the same suspicion.

Initials, former names, suffixes and all-capitals captions in a name search

A name on a legal page is not always the name as you would say it. Search for the surname alone, the reversed form, initials, earlier names, and the entity with and without its suffix, and search without case sensitivity so that a caption in capitals is caught.

Work through the forms one at a time.

• Reversed order. An index, a service list or a medical record may print "BRENNECKE, ILSE M." A phrase search for "Ilse Brennecke" misses every one of them. Search for the surname alone and narrow by reading.
• Initials. "I.M. Brennecke", "I. M. Brennecke" and "IM Brennecke" differ by spaces and full stops, and recognition can drop or add either. The surname alone covers all three.
• Former and alternate names. A maiden name, a married name, a hyphenated double surname written as two words, a nickname in emails. Ask the client; the file will not tell you a name it never uses.
• Suffixes and titles. Jr., Sr., III, M.D., Esq. A roman numeral suffix is made of the letter I, one of the shapes that can be stored as a lower case l or a digit 1, so do not make it part of the search term.
• Entity designators. LLC, L.L.C., Inc., Incorporated, Co., and the ampersand written as "and". Search for the distinctive words of the entity name and leave the designator off.
• Role labels. After the first page, a contract or pleading may never use the name again and refer only to "Tenant", "Defendant" or "the Guarantor". Confirm how the term is defined in your own document, then search for the label as well.
• Line breaks. A long name split over two lines, with or without a hyphen, is stored in two pieces. The fragment search catches each half.

Case and whole-word options are worth setting on purpose. Adobe's guide to searching PDFs in Acrobat describes a "Case sensitive" option that "searches for words with matching capitalization", and a "Whole words" option that excludes partial matches. For a name search, leave case sensitivity off. The same page lists a search preference, "Ignore Diacritics And Accents", under which "typing cafe finds both cafe and café"; if your name carries an accent, check how your reader is set. The section on exact match, case, whole words and phrases in PDF search covers these options in full.

Do wildcards and fuzzy search help find a misread name?

They help where your software has them, and what each program offers differs, so read its documentation and test it on a misreading you have already found. A plain find box in a PDF reader matches the characters you type. Wildcards and fuzzy matching are features to look for in indexing and review tools.

What one PDF reader documents

Adobe's search guide for Acrobat describes whole-word and case options, exact phrase matching, and "Match Any Of The Words". It says Boolean operators work only when you search multiple PDFs or PDF indexes, and it describes a proximity option that is available only for a search of multiple documents or index definition files and only when "Match All Of The Words" is selected. On wildcards it says: "You cannot run wildcard searches using asterisks (*) or question marks (?) when searching PDF indexes." Its stemming option is described in these words: "Finds words that contain part (the stem) of the specified search word." The example Adobe gives is a search for "opening" that finds "open", "opened", "opens" and "openly". That is a tool for word endings, not for a surname with a wrong letter in the middle.

So in a reader like that, the variant list is the tool. Adobe describes "Match Any Of The Words" as a search for "at least one of the words typed", which lets you put several one-word variants in one search.

What one indexing product documents

Dedicated search software can go further. The dtSearch feature list says its wildcard support "allows ? to hold a single letter place, and * to hold multiple letter places". It describes fuzzy searching that "adjusts from 0 to 10 so you can fine-tune fuzziness to the level of OCR or typographical errors in your files", with the example that a search for "alphabet with a fuzziness of 1 would find alphaqet". It also lists phonic searching, which "finds words that sound alike, like Smythe in a search for Smith". If your firm has a review platform or an indexing tool, look up the equivalent features in its own help.

How to test either feature

Take a misreading you harvested from the file, such as "Brermecke" on page 644.

1. Run the fuzzy or wildcard search for the correct spelling.
2. Check whether page 644 is in the results.
3. If it is not, raise the setting one step and try again, and note the setting that first catches it.
4. Look at how much noise that setting adds. A short name such as "Rill" at a loose setting will match "bill", "will" and "till" on every page.

A wildcard helps when you know where the damage is: Bre*ecke covers any reading of the double n. A fuzzy search helps when you do not. Neither replaces the list, because a name split by a space, or one that was never recognized as text at all, matches no pattern.

Finding a name by what sits next to it: address, title or account number

Search for the things that appear beside the name, then read those pages for the name itself. An address, a job title, a policy number or an email domain is a second route to the same page, and it fails independently of the name.

Pick context terms that are distinctive and that the file is likely to print wherever the person or company appears.

For Ilse Brennecke, that might be "Quarry Road", "Managing Member", the last four digits of an account number and "brh-haulage". Each hit page gets a look. Where the page shows the name and your name searches did not find it, copy what the text layer holds there and add it to the variant list. This is how the list improves as you go.

Two cautions apply. Numbers are misread too, so search for a short run of digits you can see clearly and try the same run with a likely swap; the post on why OCR gets numbers wrong lists the confusions. And context finds pages about the same subject that do not name the person, so every context hit has to be read, not counted.

Where a reader offers a proximity search, the pair of name and context can be run together. Check the conditions in your reader's documentation first. Adobe's guide, for one, limits proximity in Acrobat to searches of multiple documents or index definition files.

How sure can you be that a name does not appear in a scanned file?

Not completely. Zero hits means the characters you typed are not in the text layer. It does not mean the name is not on a page, because the name may sit on a page with no text, inside handwriting, under a stamp, or stored in a form nobody listed.

How much weight a negative result can carry depends on work you can show.

1. Every page has text. Confirm it with the method in the guide to checking whether a PDF is searchable on every page. A page with no text layer can only be read by eye.
2. The searches are logged. Each term, its options, the file and the count, including the zeros.
3. Context searches were run and read.
4. The weak pages were read by a person. Handwritten pages, faxes, pages under heavy stamps and any range where misreadings clustered.
5. A known name was used as a control. Search for a name you know is in the file, in the same kinds of places. If that search misses pages you can see, your target search would have missed them too.

Then say what you did, not more. "A search of the recognized text for the listed terms returned no results, and pages 644 to 702 were read in full" is a statement you can stand behind. "The name does not appear" is a statement about every pixel in the file. The post on searching contracts sets out the same discipline for clauses, in the section on showing a clause is absent from a scanned contract.

If the file's text is too poor to search with any confidence, recognize it again before you search further. Schema Law reads an uploaded PDF, PNG or JPEG page by page and shows the result on screen, so you can compare a name against the original before exporting a searchable PDF; the site says recognition is strongest on clean print and that nothing marks a passage as uncertain, so the comparison is yours to make. The how it works section lists the steps. Where a name is wrong in a file you must keep using, the post on correcting OCR text in a PDF covers fixing the stored word so the next search finds it.