PATNEWS: How sloppy is software prior art searching at the PTO?

[Gregory Aharonian]

!19980202 How sloppy is software prior art searching at the PTO?

But first, this week's Federal Computer Week reports that Senator
John McCain introduced legislation that calls on the Congressional
Research Service to put its reports, issue briefs and other information
products on the Internet that the public could access at no charge.
Will the AIPLA/IPO contact McCain and ask him to amend his bill to force
the PTO to put the full texts of patents onto the Internet?

FULL TEXT FULL TEXT FULL TEXT FULL TEXT FULL TEXT NOWWWWWWWWWWW

====================

Recently a patent issue that struck me as being so obvious in a well
documented field that there had to be prior art. Here is the patent title
and abstract:

5,710,561
Method and apparatus for double run-length encoding of binary data
Filed: Jan. 2, 1996

A method and apparatus for losslessly compressing binary data using
a technique referred to as Double Run-Length Encoding (DRLE). DRLE
has particular application to the compression of gray-scale data
as it is being processed for printing by a laser printer or other
continuous raster scan device. DRLE records repeating patterns of
ones and zeros with little computational complexity. Compression
ratios that may be an order of magnitude or more are obtained
frequently on data that may not compress well using traditional
Run-Length Encoding (RLE). DRLE uses a sequential history of
order-pairs that denote variable-length patterns of zeros and
ones, and then encodes these patterns as they repeat themselves.

Not only has run length encoding been written about to death, but in an era
of compression and encryption, the existence of double-DES and triple-DES
makes it obvious by analogy to multiply apply such techniques. What makes
this patent even more suscipious is that it cites only two prior patents,
and NO non-patent prior art items (surprise, surprise).

So I did the dumb thing. I dailed up INSPEC, and used the following
no-brainer Boolean search query: SU: DOUBLE and SU:RUN W LENGTH W ENCODING
Lo and behold, look what I found:

TITLE: Double encoding-a technique for reducing storage requirement
of text
SOURCE: Information Systems, vol.11, no.2, p. 177-84, 1986
ABSTRACT: The authors present a new double compression scheme for
text, program source and documentation files. The first
phase of compression is a dictionary-based method that
also uses run-length encoding. The second phase of compression
is based on utilizing the distributional and correlational
properties of the output of the first phase. The technique
achieves high degrees of compression and is further enhanced
by the hardware assistance VLSI technology will eventually
offer to the realization of data compression techniques
(23 References).

Admittedly not enough to kill the patent, but given that this paper was
written ten years before the patent was filed and has 23 references,
there is a good chance the 102-killer is out there. This paper should
have been cited by someone and listed on the front page of the patent.
The PTO likes to mislead the public by telling them examiners have good
access to the major online databases, but I find it hard to believe
when I continue to see patents like this issue for which there is easily
locatable prior art. In fact, based on my search activities over the
past few years, I guess that for 1% of issued patents - at least one of
the inventors has written a paper of relevance that wasn't submitted as
prior art - a direct Rule 56 violation. But companies can get away with
these abuses because the PTO doesn't even do the automatic trivial search
- search the databases for inventor's names.

But that's not the point. The point is that Rule 56 is a joke, in its
nearly complete lack of requiring any effort on the part of the applicant
to do any searching, especially in light of the PTO's ongoing inability
to supply examiners with enough time, money and resources to search.
Dialing up to Dialog and spending a half hour typing in random two and
three word queries out of your abstract costs $100 or so. Is that an
undue burden in order to be awarded intellectual property rights?

Thus when you are issued a software patent, it only means its construction
is statutorily correct, to a high degree of probability. But with regards
to novelty and obviousness, an issued software patent means nothing. And
1724 times nothing is still nothing. And the 13,000+ software patents
that will issue in 1998 mean nothing. Which is a crime against the small
companies that increasingly will be relying on their patents to protect
them from marketing powerhouses such as Microsoft.

Greg Aharonian
Internet Patent News Service