% David Rhead % (JANET: d.rhead@uk.ac.nottingham.ccc.vme) % \begingroup \def\Blatex{{\bf L\kern-.30em\raise.3ex\hbox{\scb A}\kern-.15em\TeX}} \def\Bbibtex{{\bf B{\scb IB}\kern-.2em\TeX}} \def\ttbackslash{{\tt\char'134}} \def\bs{\tt\char'134} \def\cite#1{[#1]} \def\label#1{{\it #1}} \let\ref\relax \let\protect\relax \def\hline{\noalign{\hrule}} \def\item#1{\par\hangindent1em\hangafter1{\noindent \hbox to 1em{\hfil#1\hfil}}} \def\itemitem#1{\par\hangindent2em\hangafter1{\noindent \hbox to 2em{\quad\hfil#1\hfil}}} \def\bi{\item{$\bullet$}} \def\rtr{$\triangleright$} \def\description{\begingroup \def\item[##1]{\par\hangindent20pt{\noindent\bf##1}}}% \def\enddescription{\endgroup} \def\section#1{\medskip\goodbreak\leftline{% \let\BibTeX\Bbibtex\let\LaTeX\Blatex \bf#1}\par\noindent\ignorespaces} \def\subsection#1{\smallskip\goodbreak\leftline{$\bullet$\fib#1}\par \noindent\ignorespaces} \def\subsubsection#1{\smallskip\goodbreak\leftline{\fib#1}\noindent\ignorespaces} \title{How might \LaTeXbf3 deal} \centerline{\bf with citations and reference-lists?} \section{I Introduction} Since there are now plans for a new version of \LaTeX\ \cite{1,2}, this may be a good time to consider how a future version should deal with citations and reference-lists. My viewpoint is that of an advisor to authors who use \LaTeX\ 2.09. I am often in the position of having to decide whether \LaTeX\ acts inappropriately or whether an author is asking for something inappropriate. Often I have the impression that \LaTeX\ sometimes makes it unnecessarily difficult for people to comply with the conventions that are standard in `academic publishing'. In making the suggestions that follow, I do realise that it's easier for me to suggest that the volunteers developing \LaTeX3 might provide some new facilities than it is for a volunteer to find time to do the work! I assume that the reader is familiar with the relevant sections of Lamport's {\sl\LaTeXsl, A Document Preparation System}. \section{II Conventions to be supported} \subsection{II(i) Citation schemes} In mainstream publishing \cite{3,4,5,6,7,8,9,10,11,12}, there seem to be three basic schemes for citations and the corresponding reference-lists: {\description \item[reference by number] In this scheme, citations are normally numbered in order of first citation. In particular, `order of first citation' is used by over 300 biomedical journals \cite{8,9}, and is specified in the ISO standard \cite{10}. (Thus, in \BibTeX\ terms, the usual numbering sequence is that given by {\tt unsrt}.) The number is used as a `label' in the reference-list. \item[author-date]~There are two main forms of citation, depending on whether or not the cited author's name occurs naturally in a sentence. In the first case, the citation is of the form \hbox{`\ \dots\ (1972) \dots\ '}, whereas in the second case it is of the form \hbox{`\ \dots\ (Crane, 1972) \dots\ '}. There are no `labels' in the reference-list, which is arranged in alphabetical order of authors' surnames (with supplementary rules for `tie-breaking'). \item[short form]~The `short form' scheme is often used when citations occur in footnotes. Typically, the first citation (or the first citation in the current chapter) gives a fairly full reference but subsequent citations use a short form. The `short form' may be introduced within the first citation, or given in a table of abbreviations. There are no `labels' in the reference-list, which may be subdivided by `type of cited document'. The scheme is common in the humanities, but also seems to be used by some software-houses when referring to their own publications (see, for example, \cite{13}). \enddescription} \noindent I think it is desirable that \LaTeX\ should provide explicit support for all three citation schemes, in order to make it straightforward to achieve the effects that people routinely achieve with traditional publishing procedures. Notice that: \item{\rtr}It is not, in general, possible to convert a document from one scheme to another (for example, from `reference by number' to author-date) automatically. Some re-writing is required. \item{\rtr}The number of items of information that need to be available differ between the schemes. For a `reference by number' citation, it is only necessary to keep track of one item (the number), whereas for the other schemes it is necessary to keep track of more than one item (that is, the author and the date, or the `fairly full form' and the `short form') so that they can be used separately. \item{\rtr}Occasionally, two different schemes may be used in parallel within the same document. (For example, in \cite{13}, a software-house seems to use `short form' when citing its own publications, but author-date when citing other publications.) \noindent Thus, it seems to me that the three schemes are best regarded as logically distinct. However, within a particular scheme, there are variations of punctuation, etc., that can be regarded as matters of `house style'. For example, some journals that have adopted a `reference by number' scheme use bracketed numerals for citations, while others use superscripts. Such variations can be accommodated by differences between style files. \subsection{II(ii) Additional references} In addition to a list of `works cited', some documents have an additional reference-list that specifies `further reading' or `all sources consulted' (see \cite{5,~pp182 \& 192}), \cite{8,~p40} and \cite{8,~p22}). There needs to be some provision for typesetting such additional lists. \section{III Deficiencies in \LaTeX\ 2.09} \subsection{III(i) Citations} When used in conjunction with \BibTeX\ and |\bibliographystyle{unsrt}|, \LaTeX\ 2.09 makes a good job of organising citations (and sorting the corresponding reference-list) according to the `reference by number' scheme as required by many journals. Generally, however, \LaTeX\ 2.09 does not provide the breadth or depth of facilities needed to support the variety of requirements for mainstream publishing: \item{\rtr}It is unfortunate that the `reference by number' sequence obtained most naturally by the do-it-yourself-er \cite{1,~p73} from \LaTeX\ 2.09 (`order of appearance within {\tt thebibliography}') is unlikely to be what the do-it-yourself-er's journal-editor requires (which will usually be `order of first citation'). The do-it-yourself-er is given no warning (either in the manual or by the software) that sorting is likely to be needed. \item{\rtr}Although one can use style-options such as {\tt apalike} and {\tt aaai} to re-define |\cite| and {\tt thebibliography} for an author-date scheme, there are many obstacles placed in the user's way: \itemitem{$\star$}The existence of the style-options is not documented in \cite{1}. \itemitem{$\star$}If one finds a style-option in a (software) archive, it may need modification to produce the precise effect required. \itemitem{$\star$}It is not obvious how one should refer separately to two items (author and date) supplied via a |\bibitem| argument originally designed for one. The do-it-yourself-er might have to study style files such as {\tt aaai.sty} and {\tt bbl} files such as those produced by {\tt aaai-named.bst} to deduce how to do this. \item{\rtr}The `short form' scheme seems unsupported. \item{\rtr}At certain points in a document, an author my need to cite several works at once. It may be necessary to specify a page (or section, etc.) for each work. For instance, \cite{6,~p404} suggests references of the form `(Kelley 1896a, 10; Kelley 1896b; Kelley 1907, 3)'. This is not easy in \LaTeX\ 2.09, since |\cite|'s optional argument applies to the citation as a whole. The author cannot supply a separate `optional argument' for each work. (By contrast, {\sf EndNote}'s analogous facilities \cite{15,~p58} seem to allow each individual work to be given its own `additional text'.) \item{\rtr}It does not seem easy to use different schemes in parallel within the same document. If |\cite| and {\tt thebibliography} are defined as required for one scheme, they will usually be unsuitable for any other. \subsection{III(ii) Additional references} As stated in section \ref{II(ii)} an author may need to typeset a list of `further reading' etc., in addition to the usual list of `works cited'. If using the {\tt thebibliography} environment from one of \LaTeX\ 2.09's standard styles for such additional references, an author will be faced with the following problems: \item{\rtr}the title will be the same as that for the list of `works cited', namely `References' for {\tt article} and `Bibliography' for {\tt report} and {\tt book} \item{\rtr}the `labels' (which may be appropriate in the list of `works cited', particularly for the `reference by number' scheme) will also appear in the additional list (where they are inappropriate), because both lists use the same definition of |\bibitem| \item{\rtr}by default, the `labels' will not be unique, since the `works cited' list and the `additional references' list will both be numbered from one \item{\rtr}it will be necessary to supply dummy {\it cite-key\/}s, purely to satisfy the syntax required for a |\bibitem|. \subsection{III(iii) Other problems} The definitions of the {\tt thebibliography} environment in \LaTeX\ 2.09's standard styles: \item{\rtr}issue either a |\section*| or a |\chapter*| command, using a {\it heading} of `References'for {\tt article} and `Bibliography' for {\tt report} and {\tt book} \item{\rtr}set {\it left-head\/} and {\it right-head\/} to either `REFERENCES' or `BIBLIOGRAPHY' \item{\rtr}do not arrange for a table-of-contents entry. \noindent These definitions can cause problems when the {\it heading}, etc.\ supplied by the standard style is inappropriate, or when a table-of-contents entry is desired. Admittedly, anyone who doesn't like the standard styles is free to take copies of the style files and modify them to suit their requirements. However, I have the impression that: \item{\rtr}among people who are competent to modify style files, modification of these aspects of the standard styles is `the rule' rather than `the exception' \item{\rtr}those \LaTeX\ users who aren't particularly computer-literate find the whole business mysterious, and seek out support staff who have to modify these aspects of the style files for them. \noindent Overall, this seems to be an area where \LaTeX\ 2.09 is failing to `free people from formatting concerns to allow them to concentrate on writing'\cite{1,~p8}. An associated problem is that modified style files may no longer be compatible with standard utilities such as {\tt lablst}. \section{IV Introduction of new facilities} It has been decided \cite{2} that \LaTeX3 will be compatible with \LaTeX\ 2.09 input files. Thus, in particular, \LaTeX3 must define |\cite| and |thebibliography| so that they have the same effect on \LaTeX\ 2.09 input files as the \LaTeX\ 2.09 definitions do. This implies that it would be difficult for \LaTeX3 to (for example) define |\cite| so that there can be an optional argument for each work in a multiple citation and define |\bibitem| so that it can have an `author' argument and a `date' argument. It therefore seems best to provide duplicates of the \LaTeX\ 2.09 facilities in \LaTeX3 (for `backwards compatibility') but to attempt to provide new commands/environments in parallel so as to provide the required functionality. The new facilities would be regarded as the `normal' facilities, would be described in the body of the successor to \cite{1}, and would be the natural choice for new users. The old facilities would be regarded as `deprecated' and relegated to an appendix of the successor to \cite{1}. Thus we can have both backwards compatibility and improved facilities for the future. \section{V Division of labour} \subsection{V(i) Details needed for document `as a whole'} The three basic citation schemes mentioned in section \ref{II(i)} determine certain details of a document `as a whole'. For each citation, there must be an entry in a reference-list. Each entry in the relevant reference-list must have associated information that can be used in citations. \subsection{V(ii) Details needed for reference-list, etc.} There are a lot of other details that need to be resolved. The information within each reference-list entry will probably need formatting according to certain rules of `house style'. The information given in citations needs organising in a consistent way (particularly for the `short form' scheme). Different people may want to assemble their reference-lists in different ways. Some people may wish to `do it themselves' \cite{1,~p73} from a physical card-index, while some may prefer to use \BibTeX\ to get details from a {\tt bib} file. In some disciplines,~proprietary systems such as {\sf EndNote} \cite{15} seem popular (because they help the user to search a database for literature to cite, as well as helping the user incorporate details of the literature into a document). Researchers may also wish to incorporate material obtained by searching details held on a {\sc cd-rom}. A reference-list generally needs sorting into a particular order. Since the list may occupy several pages, I assume that any sorting is best done outside \LaTeX, either by other software (e.g., \BibTeX) or manually by the author. \subsection{V(iii) \LaTeX\ and other software} It seems best to regard the movement of text-strings (e.g., an author's surname) within the `document as a whole' as a task that is distinct from the arrangement of details within the text-strings, and to assume a `division of labour' in which the former task is performed by \LaTeX\ while the latter is performed by some other software or manually by the author. The `division of labour' between \LaTeX\ 2.09 and \BibTeX\ seems to set a good precedent. This division of labour will lead to modular software. Once the interface between a reference-list and the rest of the document has been defined,~people can use \LaTeX\ for the body of their document, but can: \item{\rtr}experiment with different software (\BibTeX, {\sf EndNote}) for formatting the details of their reference-lists \item{\rtr}enhance the other software (e.g., \BibTeX) independently of enhancements to \LaTeX. \item{\rtr}lay their reference-lists out manually if they prefer. \section{VI \LaTeX3: A possible user interface?} \subsection{VI(i) Specifications and names} If the reasoning given in sections \ref{II}, \ref{IV} and \ref{V} is accepted, consideration needs to be given to the form that new commands/environments should take in order to support the three basic citation schemes, and to provide facilities for `additional references'. In particular, it will be necessary to choose names other than |\cite|, {\tt thebibliography} and |\bibitem| (since these names will be kept for the facilities provided for compatibility with \LaTeX\ 2.09). \subsection{VI(ii) Four sets of commands/environments} It seems to me that it would probably be convenient to have three sets of commands/environments for dealing with citations and the corresponding reference-lists, each set specifically designed to implement a particular citation scheme. Having three such sets gives scope for taking proper account of the peculiarities of each scheme, without having one scheme adversely affected by the peculiarities of another. To avoid the difficulties mentioned in section \ref{III(ii)}, it might also be worth having a specific environment for `additional references'. \LaTeX3 might, for example, have commands/environments as specified in table 1. \topinsert \hsize6.5in \centerline{\offinterlineskip \vbox{\halign{\strut#\hfil&\hfil#\hfil&\hfil#\hfil&\hfil#\hfil&\hfil#\hfil\cr \hline & Citation & Environment for & Entry in \cr & & reference-list & reference-list \cr \hline Reference by number & \bs numcite & {\tt numrefs} & \bs numentry \cr \hline Author-date & \bs dcite & {\tt adrefs} & \bs adentry \cr & \bs adcite & & \cr \hline Short form & \bs firstcite & {\tt sfrefs} & \bs sfentry \cr & \bs sfcite & & \cr \hline Additional references& -- & {\tt morerefs} & \bs moreentry \cr \hline {\it Analogue at 2.09} & \bs cite & {\tt thebibliography} & \bs bibitem \cr \hline }}} \caption{Table 1: suggested specifications for commands\slash environments in \Blatex3} \bigskip \hrule height1pt depth1pt\bigskip \endinsert Here it is assumed that: \item{\rtr}|\numcite| and |\numentry| have {\it key-list} and {\it cite-key} (respectively) as their only mandatory arguments. \item{\rtr}|\dcite| and |\adcite| have {\it key-list} as argument. |\dcite| gives a citation of the form (1972), while |\adcite| gives a citation of the form (Crane, 1972). |\adentry| has three arguments: the {\it cite-key}, the author (e.g., Crane), and the date (e.g., 1972). \item{\rtr}|\firstcite| and |\sfcite| have {\it key-list} as argument. |\firstcite| gives the form of citation to be used when a work is first mentioned. |\sfcite| gives the short form to be used in subsequent citations. The arguments of |\sfentry| might include: the {\it cite-key}, the form of reference to be used at the first citation, and the short form to be used subsequently. Whereas |\numentry| and |\adentry| can `introduce' the full reference (like |\item| starts a new item \cite{1,~p166}), it may be better for |\sfentry| to have the full reference as an argument, so that it can be used as the default `form to be used at first citation'. \noindent Although it would be desirable for the successors to the `standard styles' to define facilities for all three citation schemes, other |\documentstyle|s need not define facilities for all three. For example, a journal that wants its authors to use the author-date scheme would supply a style file that only provides author-date facilities. \subsection{VI(iii) Further details} \subsubsection{Reference by number} The |\numcite| might take the form |\numcite{|{\it key-list}|}| and similarly, |\numentry| commands would be |\numentry{|{\it cite-key}|}|. Notice that, since |\numentry| is specifically designed for `reference by number', there is no need to allow an optional {\it label} argument like that for |\bibitem|. To conform to the ISO specification \cite{10}, the successors to the `standard styles' would arrange for |\numcite| to give a citation of the form (24) and for |\numentry| to give a reference-list entry of the form \description \item[{\rm 24.}] {\sc Crane, D.} {\it Invisible colleges.} \dots \enddescription Perhaps \LaTeX3 could use the {\tt aux} file to refine an initial estimate of the width of the final |\numentry|'s `label', so that the do-it-yourself-er wouldn't need to supply a {\it widest-label\/} argument. \subsubsection{Author-date} The commands |\dcite|, |\adcite| and |\adentry| might be defined to have the forms |\dcite{|{\it key-list}|}|, |\adcite{|{\it key-list}|}| and |\adentry{|{\it cite-key}|}{|{\it author}|}{|{\it date}|}|. The {\tt adrefs} environment would {\it not} have a {\it widest-label} argument, since in this scheme entries in the reference-list are unlabelled. If such a definition of |\adentry| was documented in the successor to \cite{1}, a do-it-yourself-er would be able to use the author-date system just as easily as the `reference by number' system. To conform to the ISO specification \cite{10}, the successors to the `standard styles' would arrange for |\dcite| to give a citation of the form (1972), for |\adcite| to give a citation of the form (Crane, 1972), and for |\adentry| to give a reference-list entry with no label. \subsubsection{Short form} The |\firstcite|, |\sfcite| and |\sfentry| commands might be defined as |\firstcite{|{\it key-list}|}|, |\sfcite{|{\it key-list}|}| and |\sfentry{|{\it cite-key}|}[|{\it fairly-full-form}|]{|{\it short-form}|}{|{\it full-reference}|}|. Such definitions would, in effect, automate Butcher's manual method of ensuring consistency \cite{5,~p178}. Having {\it full-reference} as an argument means that the full reference can be used as the default {\it fairly-full-form} (to be used when the work is first cited \cite{6,~p407}). The {\tt sfrefs} environment would {\em not} have a {\it widest-label} argument. The successors to the `standard styles' would arrange for |\firstcite| to produce the {\it fairly-full-form} and |\sfcite| to produce the {\it short-form}. This makes the pessimistic assumption that \LaTeX\ can not itself determine whether a citation is the `first citation' of a particular work. If someone was ingenious enough to produce code that determined whether a citation is a `first citation', {\tt\ttbackslash firstcite} would be unnecessary (except,~perhaps, for the construction of tables of abbreviations \cite{6,~p414}.) To conform to the ISO specification \cite{10}, neither |\firstcite| nor |\sfcite| would add any punctuation (but, for example, a style file that implemented the MLA conventions would have to add brackets \cite{7, ch.\ 5}). In the successors to the `standard styles', the |\sfentry| would produce an entry with no label. People producing other style files would be free to implement other conventions \cite{5,~p168}. Some works may have no bibliography or may have just a `select bibliography' \cite{5,~p168}. For such works, it will still be necessary to supply the details for use by |\firstcite| and |\sfcite|. It may therefore be worth allowing a form of {\tt sfrefs} (e.g., |\begin{sfrefs}[null]|) that holds details of cited works but does no typesetting. \subsubsection{Additional references} Since the entries in an `additional list' will not be cited as such (although an `all sources consulted' list may contain a duplicate of a cited entry in a `works cited' list), the list will be typeset without `labels'. Even in a document that uses the `reference by number' citation scheme (and so needs `labels' in the {\tt numrefs} list), there will be no `labels' for the entries in an additional list. It therefore seems likely that the {\tt morerefs} environment could be implemented as a variation of {\tt adrefs} or {\tt sfrefs}, the main changes being: \item{\rtr}a change of title (but see section \ref{VII(iii)}) \item{\rtr}absence of {\it cite-key, author, date, fairly-full-form} and {\it short-form} arguments. \subsection{VI(iv) Order within the reference-list} As stated in section \ref{V}, it is probably best to leave any sorting of the reference-list to some other software, or to the author. However, it might be possible for \LaTeX\ to provide a warning if a reference-list is obviously in the wrong order. Perhaps: \item{\rtr}although there may be no easy alternative to numbering |\numcite|s in order of appearance within {\tt numrefs} (even though `order of first citation' is usually what is required), \LaTeX\ could give a warning if a |\numcite| gave a number that exceeded the `biggest number produced by |\numcite| so far' by more than one. \item{\rtr}there could be a warning if an |\adentry| had an {\it author} whose first letter came before the first letter of the preceding |\adentry|'s {\it author} in the alphabet. \subsection{VI(v) Citation of a specific division} %\label{division} As stated in section \ref{III(i)}, provision needs to be made for the citation of a particular division (e.g.,~page, section, chapter, equation) of another work. The syntax of citation commands should not only allow several works to be cited simultaneously, but should also allow the relevant division of each work to be specified. From the author's point-of-view, there would be a variety of satisfactory ways to specify citations that are to appear as `[4,~p10; 5; 6,~p3]', e.g. {\frenchspacing \leftline{\bs numcite\char'173smith[p. 10],brown,jones[p. 3]\char'175} \leftline{\bs numcite\char'173smith, p. 10; brown; jones, p. 3\char'175} \leftline{\bs numcite\char'173smith \& p. 10; brown; jones \& p. 3\char'175} } \noindent The precise syntax would have to take account of the practicalities of programming a command that has to be able to accept pairs of arguments, where the second member of each pair is optional. Analogous facilities would be needed for author-date and `short form' citations. Incidentally, since abbreviations such as p., ch., sec., and fig.\ are common when such divisions are specified, I think that citation commands should arrange for the optional arguments to be typeset with |\frenchspacing|. \section{VII Details of reference-lists} \subsection{VII(i) Variations within mainstream practice} Although many academic and technical publications involve only a single undivided reference-list, some such publications involve: {\description \item[more than one list] This situation can arise: \itemitem{$\star$}when there is a list of `further reading' etc., as well as the list of `works cited'. This case has been covered in sections \ref{II(ii)}, \ref{III(ii)}, \ref{VI(ii)} and \ref{VI(iii)}. \itemitem{$\star$}when conference proceedings are produced, since each contribution may have its own reference-list. \itemitem{$\star$}in manuals for software. For example, in \cite{13}, a software house's own publications are introduced in the preface and cited (in effect) using a `short form' scheme, while other people's publications are listed at the end of the manual and are cited using the author-date scheme. \item[subdivisions within a list] Some reference-lists,~particularly in the humanities, are subdivided according to the source of the cited documents (see \cite{5,~p183}, \cite{6,~p425} and \cite{7,~p88}). \enddescription} In some cases, an author may wish to add explanatory paragraphs describing, for example, how material was chosen for a `select bibliography' \cite{6, fig.\ 15.11} or information about access to (document) archives \cite{6, fig.\ 15.16}. \subsection{VII(ii) \LaTeX\ 2.09} \LaTeX\ 2.09 can cope with documents that have more than one {\tt thebibliography} environment, and seems to deal satisfactorily with a situation in which some |\cite| commands are to one {\tt thebibliography} and some are to another (provided that the {\it cite-key\/}s are unique). The default effect is to give {\it label\/}s that are not unique, which will be acceptable when each contribution to a `conference proceedings' has its references numbered from one, but not if `works cited' and `additional references' are both numbered from one (see section \ref{III(ii)}). The specification of {\tt thebibliography} \cite{1,~p187} does not allow anything other than |\bibitem|s within a {\tt thebibliography} environment. Hence, it is not clear how one can introduce subheadings within a reference-list. (In practice, a |\section*| seems to work between |\bibitem|s, but I suspect that it puts \LaTeX\ 2.09 into a loop if placed before the first |\bibitem|.) Anyone trying to add explanatory paragraphs (as in \cite{6, fig.\ 15.11} and \cite{6, fig.\ 15.16}) will probably find that \LaTeX\ 2.09 objects that `{|Something's wrong--perhaps a missing \item|}'. \subsection{VII(iii) \LaTeX3} Ideally, in order to provide support for the conventions that are routine in mainstream publishing practice, \LaTeX3 should be able to cope with all the variations outlined in section \ref{VII(i)}. \subsubsection{Multiple lists} The suggestions made in section \ref{VI} would probably cater for most situations where a document has more than one reference-list. The distinction between {\tt morerefs} and the other environments for reference-lists would take care of situations where there is a list of `additional references' as well as a list of `works cited'. The distinction between the `short form' commands/environment and the other commands/environments would take care of situations where a software house uses `short form' for its own publications and some other scheme for other publications. Conference proceedings will be able to have `a reference-list for each contribution' if \LaTeX3 follows the \LaTeX\ 2.09 precedent that allows more than one {\tt thebibliography} in a document. \subsubsection{Subdivisions and explanatory paragraphs} Because the sciences have different conventions from the humanities,~people using the `reference by number' citation scheme are unlikely to want the options of subdividing their reference-list and inserting explanatory paragraphs. Therefore, it would be legitimate to say (for example) `subdivisions and explanatory paragraphs are supported within {\tt adrefs}, {\tt sfrefs} and {\tt morerefs} but not within {\tt numrefs}', if this made the programming task easier. For example, it might be convenient to implement {\tt numrefs} as a `list-making environment' (as in \LaTeX\ 2.09) but to implement the other environments in some other way. Lack of support for subdivisions and explanatory paragraphs is unlikely to matter in {\tt numrefs}; the `other way' (more like |\paragraph|,~perhaps?) might make it easier to implement support for these facilities in the other environments. \subsubsection{Other problems} %\label{reflist-scope} One approach to some of the problems mentioned in section \ref{III(iii)} is for the standard styles to define the heading for the reference-list by, for example, |\def\numrefsheading{References}|, so that anyone who wants to change the heading can do so by issuing a |\renewcommand| command somewhere before the start of their reference-list. Another approach is to work in terms of the standard publishing industry concept of `back matter' \cite{6,~p4}. Instead of having to understand where the {\it heading}, {\it left-head}, {\it right-head} and table-of-contents entry (or lack of it) originate for units such as the glossary (if any), the reference-list(s) and the index (if any) {\em separately}, an author would only have to understand how these features are treated {\em consistently} within `back matter'. Although the `back matter' approach could be used if environments such as {\tt numrefs} followed the {\tt thebibliography} precedent and issued commands such as |\chapter| or |\section| themselves, authors might find the way that {\it heading}, {\it left-head}, {\it right-head} and table-of-contents entry materialize less mysterious if it was just the same for a reference-list as for (say) a glossary. This would imply that {\tt numrefs}, {\tt adrefs}, {\tt sfrefs} and {\tt morerefs} should not issue commands like |\chapter| or |\section| themselves. As a bonus, it would then be possible for an author to insert an explanatory paragraph before the reference-list, and to arrange for subdivisions. For example, if there was a {\tt backmatter} environment within which |\chapter| was treated as defining a unit of `back matter', an author's file might contain commands such as \begintt \begin{backmatter} \chapter{Glossary} ... \chapter{References} \section{Primary sources} \begin{sfrefs} ... \end{sfrefs} \section{Secondary sources} \begin{sfrefs} ... \end{sfrefs} \chapter{Further reading} \begin{morerefs} ... \end{morerefs} \end{backmatter} \endtt \section{VIII Conclusion} \LaTeX\ has a large number of users, and potential users, who wish to produce documents that conform to the conventions that are standard in academic publishing. One element of their requirement is the need to conform to the conventions for citations and reference-lists that are usual in their disciplines. The choice for \LaTeX3 may be between: \item{\rtr}having more facilities for citations and reference-lists than \LaTeX\ 2.09, perhaps as suggested in section \ref{VI}. This would imply an increase in: \itemitem{$\star$}the amount of code needed to implement the facilities, and the guru time needed for writing the code \itemitem{$\star$}the number of pages needed, in the successor to \cite{1}, to describe the facilities -- perhaps 8 pages rather than the 2 pages in \cite{1}. \item{\rtr}no significant increase in the facilities provided for citations and reference-lists. Contrary to the idea of `freeing people from formatting concerns to allow them to concentrate on writing' \cite{1,~p8}, many authors (perhaps most authors) would be wasting time: \itemitem{$\star$}hacking at style-files \itemitem{$\star$}searching (software) archives for ready made solutions \itemitem{$\star$}taking up support staff's time in the search for advice (with the support staff in turn taking up gurus' time in their search for solutions). \noindent Moreover, most of this time would be wasted by (or on behalf of) authors who don't want anything exotic; they just want to conform to the conventions standard in traditional academic publishing. I'm inclined to think that the first option would be the lesser of the two evils. \section{Appendix: interface with \BibTeX} The preceding sections make some suggestions for a \LaTeX3 user interface that would enable the do-it-yourself-er to conform to the conventions that are usual in academic publishing. It is also necessary to consider the implications for the interface to \BibTeX. \subsection{Single reference-list} Inspecting {\tt bst} files gives me the impression that, if it was decided to adopt a scheme such as that described in section \ref{VI}, it would be fairly easy to produce new {\tt bst} files to supersede existing ones. For example, a {\tt bst} file that implemented a `reference by number' scheme would write |\numentry| commands rather than |\bibitem| commands. If {\tt bst} files were created in this way, they would be able to deal with the straightforward situation when there is a single reference-list that contains all works cited (plus, possibly, any works specified by a command like |\nocite|). \subsection{Multiple reference-lists, all with the same style} More complicated situations can arise in which a document involves several reference-lists. For example: \item{\rtr}The editor of the proceedings of a conference might want the published proceedings to have a reference-list at the end of each chapter. \item{\rtr}If, as suggested in section \ref{VII(iii)}, {\tt sfrefs} was implemented in a way that allowed a sequence of {\tt sfrefs} environments to each be preceded by a |\section| command, then, as far as \BibTeX\ is concerned, each {\tt sfrefs} environment might be a separate reference-list. In both these examples, the document would involve several reference-lists, but each reference-list would need to be typeset in a common style. I assume that the main problems would be in arranging: \item{\rtr}to have multiple {\tt bbl} files, or distinct divisions of a single {\tt bbl} file \item{\rtr}that each reference-list takes its entries from the correct {\tt bbl} file, or from the correct division of a single {\tt bbl} file. \subsection{Two reference-lists, each with a different style} People producing documents that have a second reference-list (e.g., `further reading') in addition to the list of `works cited' might want the first list typeset in one style and the second list typeset in another. (In particular, if the `reference by number' scheme is used, the first list will have `labels' but the second list will have no `labels'.) If the commands/environments suggested in section \ref{VI} were implemented, the first list would use {\tt numrefs}, {\tt adrefs} or {\tt sfrefs}, while the second list would use {\tt morerefs}. In this situation, it would be necessary to communicate to \BibTeX\ that two lists are required, but that they are to be typeset in different styles. Since the second list is to contain `works {\em not\/} cited', it will also be necessary to specify the works to be shown in the second list. If it is decided to extend the interface between \LaTeX\ and \BibTeX\ to cater for such situations, it will probably be necessary to consider defining alternatives to \LaTeX\ 2.09's |\bibliography| and |\bibliographystyle| commands, since it seems unlikely that the syntax of the \LaTeX\ 2.09 commands could be extended so as to pass the necessary information. One might, for example, consider syntax such as |\bibtexcites[|{\it cites-style}|]{|{\it bib-files}|}| and |\bibtexmore[|{\it more-style}|]{|{\it bib-files}|}{|{\it key-list}|}|. Here, it is assumed that: \item{\rtr}the |\documentstyle| would set defaults for the {\it cites-style} and {\it more-style} that are to be passed to \BibTeX, but the user can over-ride the defaults via the optional arguments to |\bibtexcites| and |\bibtexmore| \item{\rtr}|\bibtexmore|'s {\it key-list} argument would be used to specify the works to be included in the list of `additional references'. If it is not practicable to extend the \LaTeX\slash \BibTeX\ interface to cater for these situations automatically, it would presumably be a matter of some `human intervention': \item{\rtr} to prepare a {\tt bbl} file for the `additional references' \item{\rtr} to |\input| the {\tt bbl} file. \def\item#1{\par\hangindent1.5em\hangafter1{\noindent \hbox to 1.5em{#1\hfil}}} \def\bibitem#1#2{\item{{#1}}{#2}} \section{Bibliography} \bibitem{1} {\sc Leslie Lamport.} {\it \LaTeXsl: A document preparation system.} Addison-Wesley, 1986. \bibitem{2} {\sc Frank Mittelbach and Rainer Sch\"{o}pf.} With \LaTeX\ into the nineties. {\it TUGboat,} 1989, {\bf 10}, 681--690. \bibitem{3} {\it Citing publications by bibliographical references.} BS 5605. British Standards Institution, 1978. \bibitem{4} {\it Citation of unpublished documents.} BS 6371. British Standards Institution, 1983. \bibitem{5} {\sc Judith Butcher.} {\it Copy-editing.} Cambridge University Press, 1981. \bibitem{6} {\it The Chicago manual of style.} Chicago University Press, 1982. \bibitem{7} {\sc Joseph Gibaldi and Walter S. Achert.} {\it MLA handbook for writers of research papers.} Modern Language Association of America, 1988. \bibitem{8} {\sc Edward J. Huth.} {\it Medical style and format.} ISI Press, 1987. \bibitem{9} {\sc ICMJE, International Committee of Medical Journal Editors.} Uniform requirements for manuscripts submitted to biomedical journals. {\it British Medical Journal,} 1988, {\bf 296}, 401--405. \bibitem{10} {\it Documentation -- bibliographic references -- content, form and structure.} ISO 690. International Organization for Standardization, 1987. \bibitem{11} {\sc Maeve O'Connor.} {\it Editing scientific books and journals.} Pitman Medical, 1978. \bibitem{12} {\sc Gillian Page, Robert Campbell and Jack Meadows.} {\it Journal publishing: principles and practice.} Butterworths, 1987. \bibitem{13} {\sc Marija J. Noru\v{s}is.} {\it SPSS-X Introductory Statistics Guide for Release~3.} SPSS Inc., 1988. \bibitem{14} {\it Presentation of theses and dissertations.} BS 4821. British Standards Institution, 1990. \bibitem{15} {\it EndNote: A reference database and bibliography maker.} Berkeley: Niles \& Associates, Inc., 1989. \author{David Rhead} \endgroup \endinput