Biographical Case Study: Margaret Masterman

Margaret Masterman Braithwaite (1910–1986) was a pioneer in natural language processing (NLP). Trained as a linguist and philosopher, she conducted extensive research on machine translation, information retrieval, and NLP at Cambridge in the 1950s and 1960s. Masterman also engaged in broad debates around humanities’ uses of computers in this era. As her former student and literary executor Yorick Wilks notes, her aim was to “build language-processing programs that had a sound philosophical basis.”¹

Despite her innovations, Masterman is not a well-known figure in the industry today. This is in large part because she was, in many respects, ahead of her time, as Wilks notes.² She demonstrated an interest in semantic-based translation models in a period in which the majority of research efforts, following the Chomsky school of linguistics, focused on syntax. As she argued:

Despite developing the first semantic networks with Richard H. Richens, it was only in the 1990s that this became a general area of research in NLP.⁴ Contemporary computational linguists approvingly note her “recognition that ambiguity is a consequence of the flexibility and extensibility of natural language and not a defect that can be eliminated by switching to a purified language of logic” and, decades before the argument was put forward by Donna Haraway and other postmodern inflected scholars, that “analogy and metaphor are fundamental to the creation of novel uses of language in every field, especially in the most advanced areas of science.”⁵ Masterman’s elision and tentative rehabilitation in the history of NLP must be seen as part of a series of attempts to reclaim female innovations in the field of computing – whether Ada Lovelace, Father Busa’s female punch card operators, or the contributions black female mathematicians made to NASA projects during the Space Race (and popularized in the 2016 film Hidden Figures).⁶

Masterman was born into the British elite. Her father, Charles, was a politician and writer; briefly a Liberal cabinet minister before the First World War, he became director of Wellington House, the propaganda arm of the government during the war. Her mother, Lucy Lyttleton, was a poet, literary editor, and Liberal candidate for parliament in 1929. Masterman was also the great niece of Lucy Cavendish, a pioneer in women’s education, and later in her career Masterman helped to found the Cambridge college named after her great aunt and become its first vice principal.

Masterman’s own career was as a linguist and a philosopher. She studied Modern and Medieval Languages at Newnham College, Cambridge, followed by a stint as a novelist and working in the theater. Returning to Cambridge, she studied philosophy where she was a student of Richard Braithwaite (professor of Moral Sciences and fellow with Alan Turing of King’s College), whom she married in 1932. She also attended and dictated (and was likely ejected from) Wittgenstein’s seminars that eventually made up the Blue Book. His ideas surrounding language games and the picture theory of truth proved influential for her later work.

In the early 1950s, Masterman became interested in the potential of computational linguistics. The computer industry was still in its very early stages in this period. Turing had only proposed his famous test of machine intelligence in 1950 and IBM only launched its first computer in 1952 – the same year that the term software was coined. A year later “artificial intelligence” was conceived and ENIAC, the world’s first programmable general-use electronic computer, was still running. Computers were, at this point, new, experimental machines, without random access storage, and as big as a room. The first high-level programming language, FORTRAN, was still under development.

Interest in the possibilities of machine translation was growing too. In 1954 a Georgetown-IBM system, which could translate 250 words from Russian to English, was demonstrated to the media in Washington, DC. The first major collection of essays on machine (mechanical) translation was published in 1955. Hopes were extremely high; investment in MT was spurred by the Cold War and the military were major funders of academic and industry research. So too internationalist efforts to prevent a repeat of the global conflicts of previous decades were underlining the importance of cross-cultural communication. To its proponents, MT promised to erect a new “Tower of Anti-Babel.”⁷ (I talk more about this history here.)

Masterman’s intellectual engagement was relatively early and at a point at which academic and military interest in NLP and MT was growing. In 1954, the year that Turing died, Masterman founded the Cambridge Language Research Unit (CLRU). Initially an informal discussion group with ties to the Epiphany Philosophers, the unit brought together scholars from a number of fields to produce world-leading research. Early collaborators included Michael Halliday, lecturer in Chinese and later inventor of the influential systemic functional linguistics, the biologist Richard H. Richens (who had discussed with Warren Weaver the possibilities of using word stems to create MT dictionaries a few years earlier), polymath Frederick Parker-Rhodes, cognitive scientist Margaret Boden, computer scientists Roger Needham and Yorick Wilks, and computational linguists Karen Spärck Jones and Martin Kay. Over the next three decades, Masterman spearheaded collaborative research into computational engagement with language.

The impact of the CLRU is visible both in terms of influence and funding. Many of the collaborators went on to become key figures in their fields. Masterman obtained funding from the US Navy, Airforce, and National Science Foundation; Canadian and UK government bodies; and the European Commission. This was despite the unit not being officially affiliated with Cambridge University during this time and being extremely equipment-poor. Much of their early work was conducted on Hollerith punch-card machines, and only in 1963 did they acquire a tiny ICL 1202 computer.⁸

Masterman’s collaborations at the CLRU produced ideas that were very much ahead of their time – only decades later have they become adopted as core concepts in NLP. In part this was thanks to Masterman’s impatient refusal to be limited by contemporary computing capabilities. She was, as one reviewer put it, “less than enchanted by the conventional wisdom which – inevitably, at that time – set about constricting, re-arranging, and generally butter-patting language into a form with which existing technology could cope.”⁹ The ideas that she pursued were various but her aim was to build language processing programs which demonstrated a sound philosophical basis. One of her major innovations, as mentioned briefly, was her interest in semantics within a broader intellectual climate (fueled by the popularity of Chomskyian models of language) in which syntax was king. Insisting that meaning was crucial, she proposed that the thesaurus, with its structural orientation towards semantics (but notably one which modeled tangled hierarchies), might prove a useful model on which computational linguists could build. Similarly, her thinking was deeply informed by Halliday’s work on the Chinese ideogram. Her insistence too on creativity – and metaphor – as a central part of normal language usage (rather than an exception) led her to explore toy languages and poetry.

Her interest in the insights computer programmers might garner from examining poetic and rhetorical uses of language is unsurprising given her familial and personal engagements in literary culture and language’s communicative function.¹⁰ Masterman’s own awareness of the wider cultural and aesthetic implications of computational modeling is indicated by her regular contribution to literary journals. Her reflections were published regularly in various general media outlets, including in a 1964 special issue of the Times Literary Supplement (which Coetzee read with interest while he was a postgraduate student). There she argued (against respondents fearful at the perceived threat computers posed to human creativity) that a computer program was a useful tool, offering a change of perspective or a “telescope” in the sense that it could change one’s perspective on the world.¹¹

She would do more than comment. In the 1960s she collaborated with CLRU member Robin McKinnon Wood to produce a computer program that wrote haiku. For her, the point of such a program – and toy languages in general – was to encourage nonpoets, nonexperts to play with language, play being the first stage of handling the subject seriously – or indeed the start of a “new folk art.”¹² Masterman was part of the avant-garde in this respect and her thinking developed out of close contacts with the international concrete poetry movement – and the particularly active Cambridge branch that would organize the First International Exhibition of Concrete Kinetic and Phonetic Poetry (1964). Her own haiku program was also included as part of the notable 1968 Institute of Contemporary Arts “Cybernetic Serendipity” exhibition in London.¹³ Rejecting the two cultures model of society, Masterman actively experimented with producing poetry in the machine.

Many of Masterman’s ideas are central to NLP today, if not always generated from her initial findings. Her notions that context is crucial to meaning and that semantics are a vital facet of language are indicated today in general interest in the third-wave of web development, otherwise known as the “semantic web.” Masterman’s own awareness of language’s redundancy, its everyday creative deployment and the centrality of semantics, developed in part thanks to her own literary background, was a boon to her work in NLP and MT. I contend that, as the fields begin to catch up, they might do well to consider the usefulness of attending to literary writing and scholarship in their future endeavors. Moreover, as historians of computing attend more broadly to the role and contribution of female innovators and workers, Masterman’s story is one well worth examining.¹⁴

Where to now?

Literature, Talk, and Computing

Biographical Case Study: J. M. Coetzee

Natural Language Processing

A Literary Guide to Natural Language Processing

Talking Interfaces