The word copyright is in fact misleading because it actually comprises several rights. There are indeed two categories of rights: moral rights and patrimonial rights.
Moral rights belong to the author (and later to his/her heirs): "The author shall have the right to claim authorship of the work and to object to any distortion, ..." (Berne convention, Article 6bis. In some countries moral rights are perpetual (eg in France).
Patrimonial rights may be transferred by the author to third parties. They include the possibility for the copyright holder to allow reproduction and to allow communication (circulation) to the public. The term of the patrimonial rights is usually the life of the author and 70 years after his death. Extra difficulties come from the fact that the rules under which an author can transfer his/her patrimonial rights may be more or less rigid depending on which national law applies.
Patrimonial rights are mainly meant to protect publishers rather than authors. Publishers (both academic and commercial) want to make sure that their financial investments in the publication process will be worthwhile. Scholars have another point of view: they want their works to be as widely read as possible.
A balance between these needs is necessary and is, theoretically, provided by the law which limits the term of the patrimonial rights as mentioned previously. However, the 70-year term is clearly far to long for the sake of the advancement of science.
We are facing three difficulties: (a) Most academic publications (journals, seminars) did not, until recently, ask authors for copyright transfer agreements.
(b) Even when they did, the agreements cannot, under French law and say before 1995, apply to digitisation because this possibility was not offered to authors at that time.
(c) Some French serials belong to academic institutions but they have been published by commercial publishers (eg Gauthier-Villars, then Elsevier) since the very beginning. For theses serials, we need to ask both authors' agreements and the permission of the publisher (which holds copyright on formatting). On the other-hand, since the older publishing contracts did not mention the electronic edition explicitly, publishers in principle need to secure the permission of the owner of a title before they can digitise it.
There are only two legal ways to solve these difficulties. One way is to only digitise documents that already fell into public domain, say documents with publication year < 1932. This is what the Bibliothèque nationale de France / French national library does. Another way is to seek the permissions of both the authors and the publishers involved.
We decided to choose the second option, to negotiate with publishers and -- in as far as possible -- to secure the permissions of the authors (although we believe that very few authors, if any, would actually object to have their papers digitised).
We do believe that commercial publishers see a strong commercial value in the mathematical heritage. Choosing the second option is a policy that is also meant to give us extra arguments in our negotiations with publishers.
Authors who published a paper in NUMDAM serials during the last 10 years will be contacted first and we will then go backwards by 10-year steps (until 1930). Finding authors' address and getting answers will of course get more and more difficult as we go backwards.
Digitised papers for which we did not yet get the authors' agreement will be made available on the server with the provision that "The paper will be withdrawn from the server under author's request".
We hope to have enough positive answers from (the more recent) authors to make this legal approximation acceptable.
(i) Negotiation to establish publishing contracts that are more favourable to the journals (and to the mathematical community) than the previous ones. We in particular aim at establishing a 5-year moving-wall after which papers will be freely available at the journal web pages.
(ii) Political lobbying for the inclusion into the French copyright law of an exception "for the needs of teaching and research in a not-for-profit framework". The possibility of such an exception exists in the European directive but such an exception has so far not been included into the French copyright law.
With the copyright rules specific to databases, I view the inclusion of articles into large databases by the bigger publishers as a possible threat to access the mathematical heritage in the future.
As stated by J.C. Guédon: "In short, the movement toward the privatization of databanks of fundamental science that has coincided with the digitization of commercial scientific journals is opening untold new opportunities for the Elseviers of the world." (http://www.arl.org/arl/proceedings/138/guedon.html)