Middle French, occupying in dictionary terms the ground between Tobler-Lommatzsch and Huguet, has for a long time been the Cinderella of French lexicography. A major desideratum in French philology is a dictionary of this extremely rich and substantially unexplored period. Initially conceived as a traditional dictionary, circumstances have dictated that the Dictionnaire du Moyen Français (DMF) should in fact be emerging as something both quite different from, and in its way, better than a traditional dictionary.

The DMF (www.atilf.fr/dmf) is an example of what its creator, Robert Martin, calls lexicographie évolutive. A detailed description and historical account of the genesis of the DMF is available online at http://atilf.atilf.fr/gsouvay/dmf2/PresentationDMF2.pdf. The starting point is a series of lexica (published or not) of either individual texts, or, more commonly, sets of texts belonging to specified genres. These are then mined by the dictionary search engine, to produce one or more articles, all of which will be displayed simultaneously. In other words, this is not yet strictly speaking a dictionary, but it may plausibly be described as a proto-dictionary, in that it offers substantially more than a simple concordance search in an identified text corpus. The entries are accompanied by references to the lemmas of other dictionaries: AND, TL, Gdf and GdfC, FEW, TLFi. In some, but not all cases, and only of course where such an electronic possibility exists, a hyperlink via the supplied lemma of the dictionary concerned transfers the user to the relevant website (in the case, notably, of the TLFi, but also of the AND, and of Gdf, linked via http://mcworks.ovh.org/, which functions by taking the user to an image of the relevant page of the scanned dictionary itself, so it is not strictly speaking a direct link to the relevant lemma). For the FEW, clicking on the relevant link generates all the (DMF) lemmas which form part of the same word-family, so that it is possible to extract all the relevant DMF information around a given etymon. Similarly (and this, too, is an innovation), it is possible to enter the DMF via the list of lemmas of corresponding articles in the dictionaries mentioned: so, for example, there is an AND listing where clicking on flod (AND) projects the user into the DMF article flot1. DMF1 had only one entry (9 articles), flot, for what rightly became (in DMF2 and DMF2009) two separate lemmas (flot1 < FEW *flod, flot2 < FEW *flot-). The process of linking to external dictionaries is not yet perfect, and it is not yet complete: DMF drachme lists the TLF, GdfC, and the AND, but provides a link only to the TLFi and to GdfC. Amongst the links there are inevitably a number which fail to hit a target at the other end. Nevertheless, what is now available online demonstrates without doubt the usefulness of the technology, presented in an intelligent, coherent, and user-friendly way. Again, this is an aspect which can clearly be developed and extended as time, resources, and the evolution of cognate projects permit. (For example, the DEAF will no doubt in due course be linked to in this way.)

DMF1, launched in 2003, drew on 13 lexica, presenting 32,779 lemmas and, in total, 84,778 articles. That is now the historically “fixed” DMF1 and DMF2 was a complete updating. DMF2009 now replaces that. This approach has the distinct advantage that citation of the DMF is also to a fixed point in the work's development. DMF2, four years later (2007) used 17 lexica, produced nearly double the number of lemmas (60,241), and presented 117,723 articles. The proto-dictionary DMF2009 has slightly more lemmas (60,110) and 26,000 fewer articles (91,706), reflecting the synthesis involved in the move to the 2009 version. DMF also draws on a very substantial text-base (220 “intégraux” to which access is gained either via DMF1, or via Frantext, plus 460 “partiels”, unavailable to the outside world). The amount of material which this adds is striking: so for garder (10 articles in DMF1, 14 in DMF2 and DMF2009) the textbase (nearly 7m tokens) generates a frankly unmanageable 6943 attestations for the infinitive garder alone. Even the 14 articles in the DMF itself would require considerable re-working before they could be regarded as a “dictionary” entry.

The format, appearance, and indeed the (underlying) data structures, of the DMF will be familiar to anyone who has used either Frantext or for that matter the TLFi. As with those resources, the DMF exploits the flexibility inherent in the electronic medium. Thus, for example, a search in the text base requires a preliminary definition of the corpus to be interrogated, which can then be investigated by author, date, and so forth. Similarly, within the dictionary itself, an advanced search option allows for the possibility of combining (for example) part of speech and source-lexicon, or domaine (e.g.: astronomie, chasse . . .). This is far more sophisticated than a paper dictionary; and it can only get better. Doubtless, en route additional materials will be incorporated. The DMF is a fine example of what lexicographie évolutive can offer in an era when long-term projects are no longer viable, but electronic media offer scope for constantly-evolving and constantly-improved collaborative ventures of this sort. Not only because of what it offers in terms of the contents, but also because of its methodology, the DMF exemplifies the INaLF/ATILF tradition of innovation and evolution, and the successful marriage of linguistics and information technology.