XX
accomplishments of the Dutch computer builders Gerrit Blaauw and Willem van
der Poel and the programmers Aad van Wijngaarden and Edsger W. Dijkstra.
5
A complementary history of programming languages is covered in Knuth’s
‘The Early Development of Programming Languages’ [65, Ch.1]. While Carr
and Gorn are not mentioned in the main body of Knuth’s account, they are key
players in the present article.
6
The words “machine translation” and “automatic language translation” are
used interchangeably in this article.
7
In retrospect, it is safer to restrict the claim that “everybody knew each
other” to British and American practitioners. Grace Hopper’s 1978 recollection
illustrates this point:
I had absolutely no idea of what [the Swiss] Rutishauser, [the German] Zuse, or
anyone else was doing. That word had not come over. The only other country
that we knew of that was doing anything with the work was Wilkes in England.
The other information had not come across. There was little communication,
and I think no real communication with Germany until the time of ALGOL,
until our first ALGOL group went over there to work with them. I think it’s
difficult for you in a seminar here to realize a time when there wasn’t any ACM,
there wasn’t any IEEE Computer Society. There was no communication. There
was no way to publish papers.
[111, p.23]
8
That “robot”, which was of a “certain formal character”, could deduce “any
legitimate conclusion from a finite set of premises”. Weaver adapted McCulloch
and Pitts’s theorem to mean that “insofar as written language is an expression
of logical character”, the problem of machine translation is at least solvable in
a formal sense [70, p.22].
Neither Weaver nor McCulloch and Pitts explicitly referred to Turing’s pa-
pers. Moreover, I have given no evidence here to suggest that Weaver was, by
1949, well versed in Turing’s theoretical work. Likewise, McCulloch and Pitts’s
sole, brief, and inadequate reference to a “Turing machine” in one of their con-
cluding remarks suggests that they were not well versed in Turing’s theory of
computation either. Their sole reference to a “Turing machine” was, strictly
speaking, incorrect. In their words:
[E]very net, if furnished with a tape, scanners connected to afferents, and
suitable efferents to perform the necessary motor-operations, can compute only
such numbers as can a [universal] Turing machine [74].
Nevertheless, an indirect link between Weaver’s research agenda on machine
translation and Turing’s 1936 notion of a universal machine had been established
by 1949, even though Turing’s work had yet to really surface in both linguistics
XXI
and computing. For example, no reference was made to Turing, Post, and the
like in the comprehensive 1955 book Machine Translation of Languages [70].
9
For a good understanding of the connection between McCulloch & Pitts’s
1943 paper and von Neumann’s work on the EDVAC, see Akera [8, p.118–119] and
Burks [23, p.188] in conjunction.
10
By 1946 Turing had become well aware of the fact that his 1936 notion of a
universal machine could, essentially, serve as a mathematical model of a general
purpose computer [60, p.4,6][95, p.76]. Turing, von Neumann, and their close
associates may well have been the only people to have seen this connection during
the 1940s. It was by presenting his 1950 paper, in which he devoted a section
to ‘The Universality of Digital Computers’, that Turing gradually changed the
common perception among some of his contemporaries, including Wilkes and
Oettinger [95, p.79,84][94, p.152–153].
During the 1950s, a continually increasing number of computer practitioners
began to view a universal Turing machine as a mathematical model for a general-
purpose computer that was based on the principle of a large store containing
both numbers and instructions. The practitioners included computer designers,
switching theorists, and logicians — such as Willem van der Poel, Edward F.
Moore, and Hao Wang [41, Ch.2]. Moore, for example, wrote in 1952 that a
universal Turing machine “can, loosely speaking, be interpreted as a completely
general-purpose digital computer” [78, p.51].
11
Bar-Hillel was moreover coming to the conclusion that “even machines with
learning capabilities [. . .] will not be able to become fully autonomous, high-
quality translators” [14, p.9].
12
Likewise, no mention of Turing was made in the comprehensive 1971 book
Computer Structures: Readings and Examples [16] (— I thank David L. Parnas
for bringing this book to my attention). Turing’s involvement with computer
building was popularized later, by Randell (1973), Hodges (1983), Robinson
(1994), Davis (2000), Dyson (2012), and others [97, 59, 99, 37, 48]. David Kahn’s
1967 book The Codebreakers [62] did not cover much about Turing either (— I
thank Vinton G. Cerf for bringing this book to my attention.)
13
Contrast this with George Dyson’s claim that Booth saw von Neumann’s
computer project at Princeton as the practical implementation of both Babbage’s
and Turing’s ideas [48, p.132]. For further scrutiny of Dyson’s book Turing’s
Cathedral, I refer to my review [42].
14
In modern terminology: Most computer designers who considered using a
small instruction set did not do this in connection with practically realizing a
universal Turing machine. An exception in this regard, other than Turing himself,
was the successful Dutch computer builder Willem van der Poel. In his 1952
design of his ZERO computer, van der Poel did refer to Turing’s theoretical 1936