shellreq@uklirb (Bernd Bachmann) (04/06/89)
Incremental REFER-UUCP Bibliography on Expert System Shells (IRUBESS) Pmail: H. Boley, FB Inform., Univ. 675 Kaiserslautern, Box 3049, W. Germany Email: (submissions:) shell@uklirb.uucp (requests:) shellreq@uklirb.uucp General-purpose tools made for knowledge representation & processing in expert systems (rather than special-purpose tools derived from expert systems for diagnosis, configuration etc.) are the 'shells' to be emphasized here. Such expert system shells can be regarded as -- often hybrid -- AI languages with good interactive environments because the label 'expert systems' is popularly used for a large subset of AI applications. While we are far away from 'an exact science of shells' (with subfields like 'shell semantics', 'shell correctness', and 'shell complexity'), the compilation of a bibliography on serious shell papers could help users, developers, and analysts of these -- often commercial -- tools. Because the IRUBESS data are provided in machine-readable form in a non-profit fashion, they may contribute to saving duplication of effort, which should also be attempted in other fields inside or outside AI. Since REFER is now a de facto standard for notating bibliographies in ASCII (with conversion routines to other, possibly non-ASCII, notations being simple), it was selected for the shell references in IRUBESS. Because of REFER's UUCP-wide use IRUBESS can also be made available and further expanded electronically via UUCP (email and newsgroup). We can profit from the REFER convention to sort a bibliography on demand only, just appending new entries at the end of the ASCII file that constitutes the bibliography (addbib irubess): IRUBESS can be UUCPed incrementally in well-defined parts from update to update, and recipients can simply copy these together and then have the new IRUBESS file sorted before pretty printing (sortbib irubess | roffbib | more). If you haven't used the family of REFER programs before, e.g. for collecting the sub-bibliography of references quoted in an nroff/troff/xroff text, please consult the manual pages (man refer). You may try REFER with the entries of IRUBESS-0 (see below). Should IRUBESS grow very large, indexing with INDXBIB may become necessary, but fortunately this would still support incremental growth. Is there a REFER better than our May 1986 version? Follow these rules when emailing references to one of the above eaddresses: * Send normal updates to the "shell" address with Subject "IRUBESS update", and corrections of existing entries (if such will still become necessary) to the "shellreq" address with Subject "IRUBESS correction"; give your email address (return paths are often nasty and sometimes unusable) and possibly your "shortest unique physical/postal/paper address" (as the last resort) in the body of your email, followed by the correct/ed references in REFER syntax. * Send references about shell overviews, critiques, and user experiences, as well as references technically describing individual shells, etc., but do not send pseudo-references of commercial shell descriptions (no advertisement). * Feel free to email references of those of your own papers that fit well into the category of shells compiled in IRUBESS, even if your shell is experimental only (has no external users); you should pmail hardcopies of any such emailed references to the above paper address. * Since we will watch for quality we cannot be obliged to distribute (all) references emailed to us, or to leave references distributed in an update also in later versions of the full version of IRUBESS (such a "bibliography non-monotonicity" may be unavoidable in exploding fields, where it can be hard to distinguish short-term relevance from long-term relevance); one small aspect of quality, supported by IRUBESS, is awareness of related work. * Make sure that you don't send references already in IRUBESS, if uncertain, by requesting the most recent update(s) or the full version via email (since parallel submissions cannot be avoided the final job of duplicate elimination must be done centrally in Kaiserslautern). * Use exactly the capitalization and abbreviation conventions exemplified in the sample entries of the following IRUBESS-0 posting (for authors don't use initials if you know their full first names). * Supply complete references only, e.g. both volume and issue numbers of journal articles (the normal ADDBIB doesn't prompt for the issue number, but you can end the volume line with "\\" and then type the prompt "%N" yourself). * You may send references with optional fields like keywords and abstract, but these may be eliminated from the distributed version because of the problems of overall consistency and manageable size. Other important rules of the game are the following: * Even though every sender is urged to cross-check every single reference submitted to IRUBESS, we cannot accept any liability for errors in the IRUBESS data (in this entire text "IRUBESS data" includes all updates). * It is forbidden to exploit the IRUBESS data commercially, as by selling them as a stand-alone bibliography to third parties; it is permissible to use a REFER-generated sub-bibliography of IRUBESS even within a technical paper that is distributed commercially. * Only if there is no other way to obtain the IRUBESS data (e.g. from the friend who told you about them) will they be sent directly to a recipient via email from Kaiserslautern, the normal way being its incremental submission to this newsgroup. (Perhaps an ftp option or an automatic mail replier for Subject "IRUBESS request" may be installed later.) * The IRUBESS management in Kaiserslautern may be ended at any time, perhaps after having found another site to continue the managing job. The REFER input example in the next posting, IRUBESS-0, is based on the survey references of the shell report SR-88-22, from which this IRUBESS proposal was extracted.