John Hutchins: the Complete Publications and Machine Translation Papers Bibliography

The survey literature on the history of automatic translation, recovered document by document and kept at the addresses that encyclopaedias, proceedings and reference works already cite.

A fanned stack of thin translucent sheets overlapping at slight angles

W. John Hutchins (1939–2021) was a librarian at the University of East Anglia who became the field's historian. The publications of John Hutchins are the standard secondary source on how machine translation developed: where an encyclopaedia entry on the Georgetown experiment cites a paper, or a conference tutorial footnotes the ALPAC aftermath, this is usually the work being cited. What follows is a bibliography of machine translation papers reproducing that body of work.

How the bibliography is organised

A companion contents page, machine and manual translation, covers the comparison that most of this work returns to.

Each entry keeps the address it was published at, and opens the document itself. The same argument is often revisited across two decades in different venues, so a bibliography of machine translation papers rewards reading across the years rather than down a single one: the conference version and the journal version of one argument sit several entries apart. Where a paper is named in the literature but not held here, its address resolves to the account of its subject rather than breaking.

The papers held here

How the reproductions are handled

Each paper is held at the address it was published at, in the form it was published in. Where a paper appeared twice — a conference version and a revised journal version, which happened often in this field — both are listed and the difference is noted rather than silently collapsed into whichever came later. Where only an abstract survives, the entry says so instead of implying a full text.

Why these machine translation papers still get cited

Most technical work ages out. Historical survey work does not, because later writers need a reliable account of what happened and when, and few people went back to the primary documents of the nineteen-fifties. These machine translation papers reconstructed the Georgetown demonstration from press releases and technical reports, traced the Weaver memorandum's circulation, and read the ALPAC report against the systems it was judging.

Using the publications of John Hutchins as sources

Every claim on this site that concerns dates, vocabulary sizes, rule counts or funding figures traces to one of these documents or to the primary source it quotes. Where the publications of John Hutchins disagree with a popular account, the popular account is usually a compression of a press release rather than of a paper.

What the survey literature actually established

The reconstruction work behind these papers is easy to underestimate. Establishing what the 1954 system contained meant reading the IBM release, the Georgetown technical report and the press coverage against each other and noticing where they disagree. Establishing how the 1949 memorandum circulated meant tracing correspondence rather than citations. Establishing what the committee of 1966 actually measured meant reading appendices that almost nobody had read since.

That is why these particular papers outlived the technical literature around them. A paper describing how a nineteen-eighties transfer system worked is of interest to specialists; a paper establishing what happened and when is of interest to everyone who writes about the field afterwards, which turns out to be a much larger group.

What a bibliography like this is for

Secondary literature has a shorter half-life than the documents it describes, and historical survey work is the exception. These papers stay in use because later writers need a reliable chronology and almost nobody goes back to the memoranda, committee reports and trade catalogues of the nineteen-fifties and sixties. Where a modern article gives a date, a rule count or a vocabulary size for the early systems, it is usually quoting one of these surveys rather than a primary source.

That dependence is also the risk. A single error in a widely cited survey propagates silently, because each new citation is made against the survey rather than against the document behind it. Keeping the papers reachable at stable addresses is what makes the chain checkable: a reader who doubts a figure can follow it back one step rather than accepting it.

That is also why the list below gives every document its own address rather than collecting them behind a summary. Several of these arguments were made, revised and restated across two decades in different venues, and only the documents themselves show the revisions.

Who was W. John Hutchins?

A librarian at the University of East Anglia and the principal historian of machine translation. He compiled the standard surveys of the field and the catalogue of commercial translation systems, and edited the newsletter of the International Association for Machine Translation.

Is this the complete bibliography?

It reproduces the published papers, surveys and conference contributions held in this archive. Where a paper exists only behind a publisher's paywall it is listed with its venue rather than reproduced.

Can these papers be cited directly?

Yes, and that is the point of keeping the addresses stable. Where the document itself is held here, a citation made against its address in 2008 still opens it. Where the document is not held, the address resolves to the account of its subject rather than breaking, so an old citation still leads somewhere it can be checked.

Why reproduce the papers rather than link to publishers?

Because the links break. Much of this literature appeared in conference proceedings and newsletters whose hosts have since closed, and several of the venues listed here no longer exist in any form. Where a publisher still serves a paper, the venue is named so the original can be cited.