Nature: How much content can our robots access?

In this blog ( Copyrighted Data: replies [1], Wiley and eMolecules: unacceptable; an explanation would be welcome –[2] , and elsewhere we have been discussing the “copyright” of factual information, or “data”. In [2] I ask a major publisher whether copyright applies to some or all of the factual scientific record they publish. So far I have had no reply. Here I ask another, Nature, who – at least through Timo Hannay – have been very helpful in discussing aspects of publication (most other publishers have been silent).
The issue arises in “supplemental data” or “supporting information” which is the factual record of the experiment – increasingly required as proof of correctness. Some major publishers (Royal Soc Chemistry, Int. Union of Crystallography, Nature) do not claim copyright over this; others such as American Chemical Society and Angewandte Chemie (Wiley) appear to do so, though I haven’t had a definitive public statement from either.
Our vision for the future is that a large part of published scientific data could be made directly machine-understandable, if the publishers collaborate in this. This would mean that we would have semantic knowledge accessible through engines such as Google, Metaweb, OpenLink, etc. We have built technology (OSCAR) that can extract 80% of the chemistry from scientific publications and it could index the whole literature in a few days using systems such as Condor. I heard yesterday of an image recognition system which can scan Flicker for photos of “family of four”, “red flower”, “cat and dog”, for example (feature recognition, not tagging). It looks straightforward to ask for papers which contain images of “gel”, “protein surface (GRASP)”, “aligned sequences”, “dose response curve”, “chemical formulae” etc. These are sufficiently stereotyped that I am sure this is possible – and even if it isn’t we should try.
But many publishers will simply forbid us to do this. Wiley claims graphs are their copyright. I expect they do this for gels, protein images, etc. If I ask I expect the average publisher will say you have to apply separately for permission for every image. This, of course, is impossible for a robot – we want to index a million images in a day.
The information itself is not copyright. If I sit down with a keyboard I can retype all the factual information into abstracts, collections, etc. But only the stuff that can be entered on a keyboard. Not images. Not graphs. Those I have to disassemble into words or numbers, which – in the C21 is grotesque.
So I am going to ask Nature what I can do and what I can’t. What my robots can do and what the can’t. If the answer is not “YES” to a question it is “NO” – there can be no “middle ground” for robots. If you don’t know then the answer is NO. If I have to ask for permission the answer is NO.
As background I want to praise Timo and colleagues for their support for us over the years – they have funded a summer student, and also given us an XML corpus for our SciBorg project (on which they also advise). They understand the vision.
This is already a long post, so the details and questions will be in the next one.
I’ll be using the new – and I think exciting and valuable – Nature Protocols as an example – in particular
which I think is currently Freely Accessible though not (in my language  and BBB) full Open Access.

This entry was posted in open issues. Bookmark the permalink.

One Response to Nature: How much content can our robots access?

  1. Tobias Kind says:

    Hi Peter,
    making chemistry data machine-readable is not the business of the publisher! It’s the business of the chemists themselves and it should be a requirement from editorial boards and reviewers. If chemists have to submit molecular structures and chemical property data before publication (a common fact for modern life sciences – compared to old-style chemistry) there would be no need to run any hamburger to cow algorithm like OSRA, Kekule, CLiDE, ChemOCR or Oscar. Beware(!), these are all sophisticated algorithms but their use could be avoided for new publications if raw data + metadata is directly submitted to a not yet functioning international open data chemistry repository. You can check out GenBank “Many journals require submission of sequence information to a database prior to publication so that an accession number may appear in the paper.” As long we keep reviewing journals without requesting that molecular structures and metadata and spectra and molecular property data are made publicly available and as long we serve in editorial boards of journals which don’t require submission of original molecule data and other molecular property data in machine readable format its our own fault. All this will be a painful process but it will come; it’s also a process of teaching the young chemists. The upcoming ticket system for chemistry publications requiring a accession number for each publication will be nice topic for the BlueObelisk; Don’t you think so, or is that too radical for you 😉
    Kind regards
    Tobias Kind

Leave a Reply

Your email address will not be published. Required fields are marked *