Quantcast

LEGAL NEWSLINE

Friday, March 29, 2024

Visual clustering allowed Garlock to quickly sort through asbestos claims information

Garlock

CHARLOTTE, N.C. (Legal Newsline) - Garlock Sealing Technologies was able to prove that a number of asbestos plaintiffs' attorneys had been gaming the asbestos trust system at Garlock's expense in a recent bankruptcy court while utilizing a new visual classification technology.

BeyondRecognition LLC, a Memphis, Tenn.-based document technology company and pioneer in successfully deploying visual classification technology, created its program a few years ago to visually cluster a large number of documents, narrowing them down to more manageable clusters of visually similar documents.

CEO and Founder John Martin compared the program to human vision. He said if a group of people uneducated in the Russian language were sitting at a conference table with a box of documents in Russian, anyone could organize the documents into piles according to how they look with zero understanding of the content.

"That is the power and the fundamental difference in visual similarity," Martin said.

The program only cares about how a document looks. The actual language or text doesn't matter. Martin explained the program's effectiveness as bringing a bazooka to a knife fight.

"The scope of what it does goes far beyond text-based analytics," Martin said.

Vice President of Corporate Communications Joe Howie said the image-based system is critical because many of Garlock's Rule 2019 exhibits were scanned documents with no associated text.

BeyondRecognition was able to use this technology to help Garlock discover manipulation while its bankruptcy trial was pending in a timely manner.

During Garlock's bankruptcy trial in U.S. Bankruptcy Court in the Western District of North Carolina, Judge George Hodges allowed Garlock to initiate full discovery into 15 selected cases and partial discovery into hundreds more.

Garlock found that plaintiffs' lawyers withheld evidence of asbestos exposure to products other than Garlock's gaskets and delayed filing claims with bankruptcy trusts until after obtaining inflated recoveries from Garlock in the civil justice system, revealing "a startling pattern of misrepresentation," Hodges wrote in his ruling.

Prior to Garlock's bankruptcy trial, it filed motions, requesting copies of the exhibits to Rule 2019 filings in 12 asbestos manufacturers' bankruptcy cases in the Western District of Pennsylvania and the District of Delaware in January 2011. It is unclear if those 12 bankruptcy cases were part of the 15 Garlock brought to prove manipulation.

Then in October 2011, Bankruptcy Judge Judith Fitzgerald of the Western District of Pennsylvania, who presided over the Pennsylvania cases and was designated to sit on the Delaware cases, denied Garlock's motions. Garlock appealed and the district courts reversed Fitzgerald's ruling.

However, all claimants' social security numbers had to be redacted prior to production, according to federal bankruptcy rules. The court did not have the sufficient personnel to perform the redaction so a special master was appointed to oversee the process.

Garlock had more than 200,000 exhibits on more than 3,500 CDs to sort through in less than three months if the company had any hopes of completing the task in time for the estimation trial.

To redact the social security numbers manually, where contracted attorneys read the documents page-by-page, would have required tens of thousands of man hours - resulting in roughly 20 redactions per reviewer per hour.

So the special master handling the redaction process turned to BeyondRecognition and its visual classification technology. Using the program, all 200,000 documents were automatically grouped into about 800 clusters according to their visual appearance, resulting in approximately 412,000 redactions. That kind of accuracy would have taken about "10 person years" to do manually, the company says.

According to this approach, when the technology recognizes a pattern in a document, all other documents in that cluster will follow that pattern. Attorneys then know to look at everything in those target clusters.

In the end, the technology sifts through 100 percent of the documents and attorneys only have to look through about 1 percent of those they started with to know what document types are in each cluster.

Martin compared BeyondRecognition's approach to a game of Battleship. Older programs are playing Battleship by taking shots at documents and missing, hoping to eventually get a hit. Beyond Recognition is peeking at its opponents' board, "so we know exactly where the battleships are," he said.

Other companies producing cluster programs have approached it from the standpoint of text recognition, but 40 percent of documents don't have words or sentence structure. Those documents containing "dark data" may include graphs, charts, photos or scanned documents in TIFF or PDF form.

Martin said text-based programs could cluster a patent document with a pizza recipe because they contain some of the same words.

The text string searching programs are weaker, Howie explained, because they "often have to rely on optical character recognition technology that occassionally can't convert some of the document images," or may substitutes a letter for a number or vice versa.

For example, a capital "O" could be mistaken for a zero or a lowercase "l" for a one. However, the visual classification technology returns to an old paradigm focused on appearances only "that has been completely lost in modern technology," he says.

"By using the tech. you've identified all of them, so it knows the location of each character of the document set and it can precisely redact that part of it, whether it's text or not. So BR uses an iterative approach to find social security numbers. We can use it to help find the things we are looking for," Howie said.

Martin and Howie explained how the technology can improve the court system's effectiveness.

"Any technology that improves accuracy, speed and makes the court run more efficiently should be a good thing," Martin said. "The fact that the court would run better should be in everybody's best interest."

This is especially true for asbestos litigation, where time is very important and the clock is ticking quicker for some mesothelioma claimants.

"A lot of times, especially in asbestos cases, it comes down to how long you can afford to be in litigation," Martin said.

The visual classification technology allows for a thorough, rapid job of redacting information, the company says. By saving time, cases could be over months or even years earlier.

"It could restore balance to the whole system," Howie said.

BeyondRecognition also has scanning technology that could be available and beneficial if the Furthering Asbestos Claims Transparency Act, FACT Act, were signed into law.

"If a process like that used in the Delaware and Western District of Pennsylvania could be used to obtain information from all the asbestos bankruptcies, we could have that processed in a matter of months," Howie said. "In the time that it's going to take a bill to get through the Senate and the House, we could have it done.

"Whether the bankruptcy information was obtained under FACT or via a Garlock-type protocol, BR could process everything in just a few months because we could build on the work already done in Garlock. Once all the claim information was processed it would be relatively easy to identify where plaintiffs had made incomplete or inconsistent claims, and which were the most meritorious."

The larger the mound of data is, the more effective BeyondRecognition becomes, the company says. With more documents to sift through, more redactions per hour happen. Therefore, the labor costs per document go down as the collection goes up.

"Everything this does is a force multiplier," Martin said.

For Garlock specifically, the company says it was able to perform the redactions in significantly less time for approximately $1 million less in price.

From Legal Newsline: Reach Heather Isringhausen Gvillo at asbestos@legalnewsline.com

More News