Page 10 - EDiscovery

S10 | MONDAY, MARCH 16, 2015 | E-Discovery
| NYLJ.COM

When Considering TAR,

It’s Never Too Late

Even midway through discovery,

time and money can be saved.

K
TOC
IGS
B

for patent infringement by a major brand- start work on the document collection while process. Predict uses Continuous Active
BY STEVEN M. AMUNDSON name pharmaceutical company. The plaintiff our client considered our recommendation. Learning (CAL), a machine learning protocol
AND MARK NOEL
claimed that our client’s generic products So without the beneit of TAR, we launched that has been shown in recent independent

T infringed its patents.
into a manual, linear review of the documents. studies to consistently outperform older,
echnology-assisted review (sometimes Our firm had considered TAR in other Just before we reached the midway point in more common TAR protocols. See, e.g.,
referred to as “predictive coding”) is cases, and recommended its use here. Only our review, we received approval to begin
Gordon V. Cormack and Maura R. Gross-
earning a well-deserved reputation for
after manually reviewing nearly half the col- using TAR.
man, “Evaluation of Machine Learning Pro-
its ability to reduce the time and cost of lection, however, did the irm receive approval At that point, we had already reviewed
tocols for Technology-Assisted Review in
e-discovery review in complex legal matters. to proceed with TAR. Even that late in the some 18,200 of the 40,800 total documents. Electronic Discovery, in Proceedings of the
What many lawyers may not realize, how- game, TAR produced substantial savings in Had we started using TAR at the outset of 37th International ACM SIGIR Conference on
ever, is that it is almost never too late for time and cost.
the review, we might have hoped to avoid Research and Development in Information
TAR. This is especially true of the newer “TAR reviewing even that many documents in total. Retrieval” (July 2014).

2.0” systems that are faster, more lexible Starting With Linear Review
However, those 18,200 documents gave us In practice, this means that there are not
and practical for a wide range of cases. Even the advantage of providing a ready-made set separate worklow phases for training the
when started midway through a relatively In this case, the total document collec- of seed documents to use to train the TAR TAR system and for review. All documents
small, technical review, TAR’s impact can tion to be reviewed (after applying agreed-to algorithm.
that already have attorney decisions on
still be dramatic.
search terms and culling) numbered about them are fed into the system at the start,
We saw this irsthand in a recent case in 40,800 records. While this was not a huge Running the TAR Algorithm
and the entire population is analyzed and
which we represented a generic pharma- collection by the standards of some cases, ranked—a process that runs in the back-
ceutical manufacturer that had been sued
it was nevertheless a lot of documents to get After training the system, we ran the TAR ground and takes about seven minutes for

through and would be a signiicant expense algorithm against the remaining documents a million documents.
for our client.
in the collection. We used Insight Predict, After that irst ranking is complete, the
STEVEN M. AMUNDSON is a partner with Frommer Believing that TAR would enable us to get the TAR platform developed by Catalyst, system provides a small batch of docu-
Lawrence & Haug. MARK NOEL, a former IP litiga- through the review more quickly and there- which has the ability to use any and all ments to reviewers that mostly contains
tor, is managing director of professional services at fore at less cost, we recommended it to our cli- previously coded documents (called “judg- the next-best, unreviewed documents the
Catalyst.
ent. But looming deadlines demanded that we
mental seeds” in TAR parlance) to start the
system can ind, but also includes a few