Corpora: Re: US company claims patent on common machine translation

Bill Imlah (imlah@cre.canon.co.uk)
Mon, 13 Oct 1997 13:20:15 +0100

Jacques Guy wrote:
>
> A post appeared the other day on sci.lang and soc.culture.esperanto,
> the import of which ought to be of utmost concern to recipients of this
> list.
> In short, some tiny US company is claiming patent on a very common
> approach to machine translation.

I'm not a patent agent (so these comments are just my understanding
of the situation, and not a definitive guide) and I'm not fully
familiar with the patent that you refer to, but, as far as I
understand, there are several escape routes available if you are
sued for alleged infringement of a patent. For instance you might
challenge its validity:

Lack of novelty: you might be able to show that the invention
was already known at the priority date, ie that it was already
in the public domain, eg. by reference to documentation that
existed before the patent's priority date and that the patent
examiner didn't know of when the patent was granted.

Obviousness: you might argue that what the patent describes
was obvious at the time. For instance, if there already
existed publicly-known translation methods that used a natural
language (eg Russian) as an interlingua, and ones which
used constructed languages (eg semantic representations) as
an interlingua then you might want to argue that what
is described is obvious, so that there is no inventive step.
[obviousness in retrospect can be difficult to prove, so it
would help if you can find prior publicly-available documentation
that relates to this or similar ideas.]

Insufficient disclosure: the specification should be sufficient
that a person skilled in the art can understand and work the
invention (given the known state-of-the-art at the time).
Failure in this area can cut down the scope and usefulness
of a patent. For instance, you say the main claim states
"there being no requirement of reworking of said intermediate
pathway language to accommodate changes in source or
target languages", so you might want to argue that the patent
does not disclose enough information to make it clear to
someone skilled in the art how to build a system that can do
what is claimed.

You might also argue that the invention as claimed does not cover
your system. For instance, the main claim of the patent includes
the text "including a first substep that parses said source text
into individual sentences, with each sentence in its own separate
subfile". If your system does not include a step of parsing text
into individual sentences each in its own separate subfile then,
depending on what this means in the context of the patent spec,
you might want to argue that you don't infringe in the first
place.

All this is just my personal understanding, of course, and I do
not know enough about your system, patent law, or the patent you
refer to to comment on your specific situation, so please do not
rely on any of this other than to suggest possible avenues for you
to explore. For reliable advice, your best option is probably a
patent attorney (or, less expensively, you may find that your
patent office is a good first place to get advice).

Bill

-- Bill Imlah imlah@cre.canon.co.uk
Canon Research Centre Europe, 20 Alan Turing Rd, Guildford,
Surrey, GU2 5YF, UK. http://www.cre.canon.co.uk/
voice: +44-1483-448844 fax: +44-1483-448845