III
into applied linguists and many professional linguists became programmers [14,
p.7][70, p.2,3,12,24][90, p.207]. Documenting this technological convergence leads
to a better understanding of when, how, and why Turing assumed the mantle of
“the father of computer science”.
In this article, I describe the convergence by zooming in on the work of
Booth, Carr, Gorn, and Oettinger.
4
I will show that, not until the 1950s, did
computer programmers like Booth and Gorn begin to reinterpret the electronic
computer in terms of the universal Turing machine. They did this with the pur-
pose of developing higher level programming languages.
5
Turing thus assumed
the mantle of “the father of computer science” for reasons that are orthogonal
to the commonly held belief that he played an influential role in the design or
construction of “universal computers”. My historical account is primarily about
the 1950s and ends with a brief discussion of Turing’s 100th birthday in 2012
and with published work related to this article.
The take-away message in general terms is that the 1950s constitute a decade
of cross fertilization between linguistics, computer programming, and logic. That
decade is preferably not viewed as a smooth road from modern logic to comput-
ing; if there was any road at all in the history of computing, then it was most
definitely from practice to theory.
2
Machine Translation — a Bird’s Eye View
“See what you can do with your Russian” — the student Oettinger was told
around 1949 by the American computer pioneer Howard Aiken at Harvard af-
ter the latter had corresponded with Weaver on the vexing topic of machine
translation.
6
Weaver had directed American war work of hundreds of mathe-
maticians in operations research. Fully aware of the developments in electronic
computing machines, he had come to believe around 1946 that code-breaking
technology from the war could help detect certain invariant properties that were
common to all languages. Weaver had expressed his ambitious ideas in writing
in his 1949 memorandum Translation [70, Ch.1] which, in turn, sparked intense
American interest, including Aiken’s interest, in the possibility of automatic lan-
guage translation. In comparison with Booth’s mechanical dictionary from 1952,
Weaver’s original idea on machine translation was more ambitious — namely to
go “deeply into the structure of languages as to come down to the level where
they exhibit common traits”. Instead of trying to directly translate Chinese
to Arabic or Russian to Portuguese, Weaver supported the idea of an indirect
route: translate from the source language into an “as yet undiscovered universal
language” and then translate from that language into the target language [70,
p.2,3,15,23].
In the academic year 1949–1950 Oettinger started thinking about mechaniz-
ing a Russian-to-English dictionary. He also stayed at Maurice Wilkes’s comput-
ing laboratory in Cambridge for a year where he met Alan Turing on a regular
basis. Wilkes, in turn, also visited leading figures in computing. He regularly
traveled from England to the United States where he met with Aiken, von Neu-
IV
mann, and many others. Apart from being an accomplished computer designer,
Wilkes was also practicing and advancing the art of computer programming. In
this regard, around 1952, he met Perlis and Carr who were working with Project
Whirlwind at MIT [113, p.31]. Perlis had obtained his Ph.D. in mathematics from
MIT in 1950. Carr had done the same in 1951 and had also spent some weeks in
Wilkes’s computer laboratory in England together with Oettinger. The world of
computer practitioners was thus a very small one: many practitioners knew each
other on a personal level and several became involved in the ACM [7, 113].
7
Initially, most linguists were rather pessimistic about Weaver’s memoran-
dum, relegating his aspirations about machine translation to the realm of the
impossible. Gradually, they started to see opportunities [70, p.4,137]. By 1951
the computer had made a clear mark on linguists. The Israeli linguist Bar-
Hillel conveyed that message by making an analogy with chemistry. Chemists,
he said, need “special books instructing students how to proceed in a fixed se-
quential order [. . . in their] attempted analysis of a given mixture” [14, p.158–159,
my emphasis]. Likewise, special books will have to be written for the linguist,
books that contain “sequential instructions for linguistic analysis, i.e., an op-
erational syntax ” [14, p.158–159, original emphasis]. According to the historian
Janet Martin-Nielsen, American linguistics at large transformed from elicitation,
recording and description before the war to theory and abstract reasoning af-
ter the war. It “rose to prominence as a strategic and independent professional
discipline” [73].
Researchers knew that literal translations would yield low quality machine
translation. Therefore, some of them sought methods to construct “learning
organs”; that is, machines that “learn” which translation to prefer in a given
context [14, p.154]. As Weaver had already put it in 1949: the “alogical elements”
in natural language, such as “intuitive sense of style” and “emotional content”,
rendered literal translation infeasible [70, p.22].
Weaver’s remarks can, in retrospect, be viewed as part of a grander intellec-
tual debate in which fundamental questions were posed such as whether machines
can think (cf. E.C. Berkeley’s Giant Brains, or Machines That Think [18]).
Weaver had addressed this issue optimistically in 1949. A year later, Turing’s
1950 article ‘Computing Machinery and Intelligence’ was published [105]. It was
followed up by Wilkes’s ‘Can Machines Think?’ [112].
Weaver’s memorandum was based on an appreciation for the theoretical 1943
work of McCulloch & Pitts, entitled ‘A Logical Calculus of the Ideas Immanent
in Nervous Activity’ [74]. McCulloch and Pitts had essentially tried to find a
mathematical model for the brain. Weaver described their main theorem as a
“more general basis” for believing that language translation was indeed mech-
anizable by means of a “robot (or computer)”.
8
In other words, according to
Weaver there was no theoretical obstacle to machine translation: learning or-
gans could, at least in principle, resolve the translation problem.
After leaving Aiken’s lab at Harvard to temporarily join Wilkes’s research
team in Cambridge, Wilkes made clear to Oettinger that he had “no use for lan-
guage translation”. Instead, he urged Oettinger to address the question whether