Highlighted Selections from:

Machine Translation: A Tale of Two Cultures


DOI: 10.1002/9781118613504.ch10

Lennon, Brian. “Machine Translation: A Tale of Two Cultures.” A Companion to Translation Studies (2014): 133–146. Print.

Summary: One product of the US–Soviet arms race was a vision of computers as fully autonomous translators of human writing and speech in natural languages. In the United States, this vision was embraced by some prominent postwar mathematicians and engineers, contested by others, and regarded with caution or dismay by most humanists and writers and many journalists. Debate over the technical and ethical limits of computing was widespread and energetic, both in the academic world and the US literary and journalistic public spheres; literature and literary language had a surprisingly prominent place in this debate, as the last frontier for the power of computation and its ultimate test. As such (and in the United States at least), the history of machine translation, or “MT,” provides a vivid illustration of the postwar conflict of what C.P. Snow called the “two cultures” of applied science and the humanities.

p.1: In this respect, at least, the postwar internationalism of early MT research can be situated squarely within a Euro-American or Euro-Atlantic intellectual tradition shaped by the historical concurrence of secularization, nationalism, and empire. -- Highlighted may 4, 2014

p.2: Mechanical or mechanizable translation methods were implied by both philosophical and practical auxiliary languages, the ideal of which was to restrict each single word to a single unambiguous meaning (thus John Hutchins [1986], for example, refers to the works of Beck, Kircher, and Becher as “mechanical dictionaries” [22]). By contrast, the “machine” in “machine translation” designates a non-human translating agent, designed to take the place of the human translator sooner or later, and ideally altogether, at least for some of the earliest researchers in the field. As in the field of artificial intelligence (AI), which like computational linguistics has its origin in early work on MT, the goal of fully automated natural-language processing – that it be suficiently accurate to pass the “Turing test” by persuasively simulating the discourse of a human being – represents the cultural power of the speculative imagination in this work: from 1949 to 1966, both enthusiasts and skeptics described fully automated high-quality translation (FAHQT) in mythic terms, as a “holy grail.” This goal structured debate across the entire field, pitting theoretical against pragmatic approaches (and optimistic and pessimistic assessments of work of each type), strongly inluencing public perception of the research, and, in time, leading to collapse and retrenchment. -- Highlighted may 4, 2014

p.2: In a 1947 paper entitled “On a Translation Machine Built on the Basis of Monolingual Language-Translation Methodology,” Troyanskii imagined a “universal logical make-up in all languages” accessible using “about 25 universal international symbols of logical parsing for all languages ... capable of rendering without exception all relations and the slightest shades of human thought” and ensuring “absolutely exact translation into other languages without distortion of meaning” (Hutchins and Lovtskii 2000, 204). -- Highlighted may 4, 2014

p.3: The idea of a logical interlingua manipulable by a machine resurfaced in the postwar writings of Warren Weaver, the mathematician and engineer who served as a director at the Rockefeller Foundation and the US Ofice of Scientiic Research and Development (OSRD) during and after the war. (Weaver seems not to have been aware of Troyanskii’s projects.) In discussions during 1946 with Andrew Donald Booth, then beginning work on the construction of computers at Birkbeck College, London, Weaver speculated about new applications for the Colossus code-breakers constructed during the war at Bletchley Park, suggesting that cryptanalytic techniques might be applied to the translation of natural languages. Weaver would pursue this approach for several years, writing in a 1947 letter to the cybernetics researcher Norbert Wiener: “When I look at an article in Russian, I say: this is really written in English, but it has been coded in some strange symbols. I will now proceed to decode” (Weaver 1955, 18). -- Highlighted may 4, 2014

p.3: The conclusion Weaver drew from this, that a logical basis for all existing languages might be accessed with cryptanalytic techniques, was very quickly discredited. Still, its basic impulse, which one might call the neutralization of culture through the segregation of soluble engineering problems from potentially insoluble philosophical ones, pervaded subsequent work in MT as a constant temptation. -- Highlighted may 4, 2014

p.3: In many ways, the story of MT is the story of an attempt to assert the independence of computation from culture and, at the same time, to assert computation’s dominion over culture: a story in which applied science played a more aggressive and destructive role in the postwar university than C. P. Snow cared to recognize, in his polemic against the division of “two cultures” (Snow 1946). -- Highlighted may 4, 2014

p.4: Weaver placed the neutralization of culture in the service of an internationalist ideal, describing the multiplicity of human languages as a “world-wide translation problem” that “impedes cultural interchange between the peoples of the earth, and is a serious deterrent to international understanding” (Weaver 1955, 15). -- Highlighted may 4, 2014

p.5: Reiler would eventually set aside his early reservations about MT as a “new expansion of the empire of the machine,” abandoning his claim for the necessity of pre and/or post-editing and declaring the full automation of translation a practically achievable goal (Reiler 1955, 136, 143). -- Highlighted may 4, 2014

p.6: William E. Bull, Charles Africa, and Daniel Teichroew cautioned that in such cases, “no translation at all would be less dangerous than a wrong or misleading one” (Bull et al. 1955, 95). -- Highlighted may 4, 2014

p.6: It was the beginning of a golden age for MT, deined by major international conferences, a critical mass of important publications, and (in the United States) easy access to generous government, military, and private funding even before the Sputnik crisis of 1957. -- Highlighted may 4, 2014

p.6: Hutchins suggests that while this inlux of funding after 1954 was driven mainly by Cold War geopolitical objectives, the cultural fascination with artiicial intelligence, both among the public and among scientists themselves, may have helped boost support for MT research as well (1986, 58–59). -- Highlighted may 4, 2014

p.6: Noam Chomsky joined the MT lab at MIT, developing work on syntax that would inluence the direction of subsequent work, though Chomsky himself would come to feel MT was “pointless” and “hopeless” (Hutchins 1986, 89, 181); -- Highlighted may 4, 2014

p.7: Some researchers suggested that MT might be applied in extending long since mechanized modes of literary study itself. Mechanical Resolution of Linguistic Problems (1958), a volume co-authored by Booth and two of his doctoral students at Birkbeck, Leonard Brandwood and J. P. Cleave, described their use of “digital calculators” in the stylistic analysis of Plato’s dialogues as venturing “like Daniel, into the den of [their] colleagues in the Faculty of Arts” (Booth et al. 1958, v). Others followed with less trepidation, triumphantly announcing a “change in the climate of opinion among literary scholars” presaging a “revolution in literary studies” (Levison 1967, 193). -- Highlighted may 4, 2014

p.9: J. C. R. Licklider, then head of the US Advanced Research Projects Agency’s Information Processing Techniques Office, had counseled IBM not to invest in MT product services (1966, 19). “Unedited machine output from scientific text,” it concluded, “is decipherable for the most part, but it is sometimes misleading and sometimes wrong (as is postedited output to a lesser extent), and it makes slow and painful reading” (1966, 19). Finally, it noted that “in some cases it might be simpler and more economical for heavy users of Russian translations to learn to read the documents in the original language,” adding that many US scientists already did just that, that instructional resources were available for those inclined to make use of them, and that acquiring basic reading facility in Russian was not likely to divert large quantities of a researcher’s time (1966, 5). -- Highlighted may 4, 2014

p.11: crude but functional Web-based MT is reflected in the literary production of pseudo-avant-gardes like the “Flarf poets” who emerged in the United States in the mid-2000s. These culturalizations of the culture of computation we have been calling “MT” certainly support Hutchins’s observation that

[t]here is now a growing realization that for many recipients stylistic refinements are not necessary; it appears that on the whole users are more content with low quality texts than translators and post-editors. (1986, 331)

But they also give it something of a twist. -- Highlighted may 4, 2014