https://d2ng68cf36kxsk.cloudfront.net/store/public/post/1233/post_images/Slide405.JPG">

After that, he went to work for IBM, where he joined a team developing a purely statistical approach to machine translation.¹⁰⁵ Instead of trying to figure out how language worked, all they had to do now was to dump already translated texts into the computer and do statistical analysis – “bags of words,” the doyens of natural language processing would call these.¹⁰⁶ (Ironically, this phrase was first coined in 1954 by Chomsky’s adviser, Zellig Harris, although then he was trying to say that language is not just that.¹⁰⁷)

“Regrettably,” said Mercer, with co-author Church, “interest in empiricism faded in the late 1950s and early 1960s with a number of significant events including Chomsky’s criticism … in Syntactic Structures.” Now they were advocating a return to empiricism, in the form of a “data-intensive” approach to language that they called “Text Analysis … a pragmatic approach that is well suited to meet the recent emphasis on numerical evaluations and concrete deliverables. Text Analysis focuses on broad (though possibly superficial) coverage of unrestricted text, rather than deep analysis of (artificially) restricted domains.”¹⁰⁸ Mercer’s boss was pleased with the results of their work, Mercer recalled in his 2014 acceptance speech for a Lifetime Achievement Award from the Association of Computational Linguistics. “Whenever I fire a linguist,” the boss said, “the system gets better.”¹⁰⁹

Statistical machine translation is what would make Google Translate work when it was launched in 2006 (at least in its first iterations^§AS2.2.1d), and work quite well. Its power was not so much the algorithms that drove it, but the size of the bilingual corpus that Google had managed to copy from the web. When a word, phrase, or sentence needs to be translated something similar can be found, and the larger the corpus, the better it works. Where theoretical structuralist linguistics had failed, Statistical Machine Translation would be a triumph of US empiricism.

---

¹⁰⁵ Peter F. Brown, John Cocke, Stephen A. Della Pietra, Vincent J. Della Pietra, Fredrick Jelinek, John D. Lafferty, Robert L. Mercer, and Paul S. Roossin, 1990, “A Statistical Approach to Machine Translation,” Computational Linguistics 16(2):79–85; Peter F. Brown, Stephen A. Della Pietra, Vincent J. Della Pietra, and Robert L. Mercer, 1991, “Word Sense Disambiguation Using Statistical Methods,” pp. 264–70 in ACL ’91 Proceedings of the 29th Annual Meeting of the Association for Computational Linguistics, Berkeley, CA.

¹⁰⁶ Helmut Schmid, 2010, “Decision Trees,” pp. 181–96 in The Handbook of Computational Linguistics and Natural Language Processing, edited by A. Clark, C. Fox, and S. Lappin, Chichester: Wiley-Blackwell, pp. 188–91.

¹⁰⁷ Zellig S. Harris, 1954, “Distributional Structure,” Word 10(2–3):146–62, p. 156.

¹⁰⁸ Kenneth W. Church and Robert L. Mercer, 1993, “Introduction to the Special Issue on Computational Linguistics Using Large Corpora,” Computational Linguistics 19(1):1–24, p. 1.

¹⁰⁹ www.bloomberg.com/news/features/2016-01-20/what-kind-of-man-spends-millions-to-elect-ted-cruz-

---

Kalantzis, Mary, and Bill Cope. 2020. Adding Sense: Context and Interest in a Grammar of Multimodal Meaning. Cambridge UK: Cambridge University Press. pp. 220–221. https://doi.org/10.1017/9781108862059

Statistical Machine Translation at IBM §AS2.2.1a