31.1 Prologue: On Visionaries and Their Visions

I initially finished this contribution on the day Sue Atkins passed away. Sue Atkins was not only a great visionary in the field of lexicography; she also changed the lives of those who came into her orbit. More than that, she made them work so hard that her futuristic insights could be transformed into actual working tools and products. This achievement should not be taken lightly. People who are ahead of their time often work in isolation or are simply not understood by their contemporaries. Who would have thought that the obscure mathematics of the theoretical physicist Paul Dirac, which had been gathering dust for decades, would be rediscovered and put to good use for detecting and recovering errors in digital computing devices (Reference Kaye, Laflamme, Mosca, Kaye, Laflamme and MoscaKaye et al. 2007)? Think: Scratch a CD-ROM and it can still be read. Listen to Ludwig van Beethoven’s Hammerklavier or his Razumovsky string quartets, knowing that this type of music utterly confused his audience 200 years ago, with musicians then finding it unplayable; yet listen to it today and find it most enjoyable. Also remember that Beethoven knew he was not bringing notes together for his contemporaries, claiming: “Oh, they are not for you, but for a later age” (Reference McCallumMcCallum 2020). Bring in Sue Atkins and realize that her most provocative concept – the “virtual dictionary” (Reference Atkins, Gellerstam, Järborg, Malmgren, Norén, Rogström and PapmehlAtkins 1996), one that exists only at the time of access (Reference de Schryverde Schryver 2003, 162–163) – is simply not for mere mortals to comprehend. As Reference Atkins, Kiefer, Kiss and PajzsAtkins (1992a, 521) rightly warned, if our future dictionaries “do not result in a product light years away from the printed dictionary, then we are evading the responsibilities of our profession.” She spoke those words at a EURALEX conference over three decades ago; since then, in my view, we have indeed been sleeping at the wheel.

Yet in her orbit, Sue Atkins worked tirelessly to bring the best linguistic theories to lexicography, notably the application of Frame Semantics to corpus analysis with the linguist Charles J. Fillmore, which resulted in a string of highly influential theoretical papers (Reference Atkins, Sue and JohnsonAtkins et al. 2003a; Fillmore and Reference Atkins, Kiefer, Kiss and PajzsAtkins 1992, Reference Fillmore, Sue Atkins, Sue Atkins and Zampolli1994, Reference Fillmore and Sue Atkins1998, and Reference Fillmore, Sue Atkins, Ravin and Leacock2000). Her collaboration with the corpus linguist Adam Kilgarriff eventually resulted in the creation of the Sketch Engine, “the de facto standard software for developing dictionaries and for corpus linguistic research” (Reference RundellRundell 2015, 460). And then we have the binary star system, two of the UK’s practical lexicography stars orbiting around some elusive barycentre, performing an elegant tango: Sue Atkins on the one hand and Patrick Hanks on the other. Through their endless teasing and nudging of one another, they moved the field forward with their talks and writings.

The “real story” of the first COBUILD dictionary, for instance, when both Atkins and Hanks were involved in the revolutionary dictionary project of Sue’s brother John McH. Sinclair, remains to be written. To this date, COBUILD 1 (Reference SinclairSinclair 1987), for which Hanks was the managing editor, remains one of Hanks’ major contributions to practical lexicography. The probability measures proposed in Hanks’ most influential paper, “Word association norms, mutual information, and lexicography” (Reference Church and HanksChurch and Hanks 1990), were the starting point for Kilgarriff’s Word Sketches, which is now a component of the Sketch Engine. Another early project, the Hector Project, was “the first systematic attempt ever to link word meaning with word use using corpus evidence” (Reference de Schryver and de Schryverde Schryver 2010a, 11). The most relevant publications for Hector appeared as another pas de deux, at successive COMPLEX conferences: Reference Atkins, Kiefer, Kiss and PajzsAtkins (1992b) and Reference Hanks, Kiefer, Kiss and PajzsHanks (1994). In his magnum opus Lexical Analysis: Norms and Exploitations, Hanks even goes as far as asking the rhetorical questions: “If Sue Atkins shakes the salt, is Sue Atkins to be regarded as some kind of a [FORCE]? […] But then, what if Sue Atkins shakes the world with her revelations about lexicography?” and he concludes with the realization that the semantics at work “cause cognitive destabilization” (Reference HanksHanks 2013, 98).

The application of Frame Semantics to lexicography, the use of the Sketch Engine in corpus-driven dictionary-making, the fruitful dialogue with fellow lexicographic giants – all of this was sped up thanks to Atkins’s association with Michael Rundell, who had also been part of the COBUILD 1 team, where they first met. Later, Atkins and Rundell teamed up to teach others about these novel concepts and new tools, first in South Africa (Grahamstown in 1997, Pretoria in 1998), later in Europe (Brighton in 2001, followed by various other locations and return visits over the next two decades), as well as in Asia (Hong Kong in 2006), Oceania (Auckland in 2012) and the Americas (Boulder, Colorado in 2016). This hugely important outreach program is now known as the “Lexicom workshop in lexicography and lexical computing” series, with Adam Kilgarriff responsible for the “lexical computing” part from 2001 until 2015. The written material first used in South Africa eventually morphed into a key publication: The Oxford Guide to Practical Lexicography (Reference Rundell and FontenelleAtkins and Rundell 2008). Himself an early believer in corpora (Reference Rundell and StockRundell and Stock 1992a, b, c), Rundell contributed significant original thinking about the future of lexicography, in solo work and with Atkins and the colleagues orbiting her (Reference 700Atkins, Rundell and HiroakiAtkins et al. 2003b; Reference Rundell and CorréardRundell 2002; Reference Kilgarriff, Husák, McAdam, Rundell, Pavel, Bernal and DeCesarisKilgarriff et al. 2008; Reference Rundell and CorréardKilgarriff and Rundell 2002; Reference RundellRundell 2014 and Reference Rundell, Hanks and Gilles-Maurice2017; Reference Rundell, Kilgarriff, Meunier, De Cock, Gilquin and PaquotRundell and Kilgarriff 2011).

I have written about the future of lexicography in a number of earlier studies. Two decades ago I wrote “Lexicographers’ dreams in the electronic-dictionary age” (Reference de Schryverde Schryver 2003), a decidedly upbeat and near-exhaustive discussion of the hot ideas in the field at the time. Dream 112 dealt with “electronic dictionaries in which the potential is explored to link an automatically derived dynamic user profile to the proffered multimedia lexicographic output” (Reference de Schryverde Schryver 2003, 189). Seeing that such types of dictionary were not being compiled in the years following the publication of “Dreams,” I grew frustrated and set out to give a fuller account of my own view of what “adaptive and intelligent dictionaries” could look like (Reference de Schryverde Schryver 2010b), using the concept of (Fuzzy) Simultaneous Feedback as a starting point (Reference de Schryverde Schryver 1999, Reference de Schryver2005, and Reference de Schryver, Gouws, Heid, Schweickard and Wiegand2013). At the eLex 2019 conference, I was interviewed for the Brazilian journal Calidoscópio and gave “An overview of digital lexicography and directions for its future” (Reference de Schryver, Chishman and Brunade Schryver et al. 2019a) – an interview in which I was neither upbeat nor frustrated, just realistic.

The problem with being realistic is that we are running after the facts, and the facts are that as lexicographers we are becoming irrelevant. To the majority of today’s youth, a dictionary means either that old book gathering dust on grandmother’s shelf or anything a search engine returns for any imaginable query, whether it be textual (running words), visual (pictures and videos), auditory (music), combinations of these (a newspaper article with photos; a music video), or all of it combined (subtitled movies). While we lexicographers were sleeping at the wheel, we were succeeded by the Big Data companies, and they seem no longer to need us. This is at the very least annoying, as one of the starting points of the development of all search engines was exactly the manually compiled reference works – the ones compiled by us lexicographers. Lexicographers were even used as “guinea pigs” (often unknowingly) to help build and test such systems. To better understand where lexicography and metalexicography come from, in the hope of saying something new and valuable about the future of dictionaries, it is instructive to briefly review the past half century.

31.2 The Past: The Last Half Century Through the Eyes of Sue Atkins

When I learned that Sue Atkins had passed away, my first instinct, immediately implemented, was to invite the entire lexicographic community to re-watch the last talk given by her; it was a recorded conversation with Michael Rundell – as filmed and edited by Izzie Reference ShawShaw (2020). While that video had been online for about a year, the number of views almost doubled overnight, to two thousand. As so well formulated by Rundell in the video, “Sue’s impact on the profession, on the way we produce dictionaries, on the way we think about language, really can’t be overestimated.” The video may be characterized as an extended TED event, marrying a lexicography masterclass, an autobiography, a review of the field from the 1970s onwards, taking in milestone papers and colleagues, while also applying small vignettes to the heavenly bodies in Sue Atkins’ orbit.

It is hard to imagine today that when Sue Atkins started off her career working for Collins, she was simply told to look at other dictionaries for inspiration, to write out her dictionary articles on six-by-six index cards (which had to be sent back to headquarters by post), to use three colors of ink (black, red and green – each for a distinct typeface), and to phone colleagues after 18:30 (for the cheaper tariff). The state of many (competing) bilingual dictionaries then was one in which translation equivalents were listed in succession, separated merely by semicolons. By the time she completed compiling her English–French dictionary (Atkins and Robert 1978), she had developed the first style guide, come up with a practical approach to dealing with register labels, designed methods to choose good examples, and learned to think really hard about how to approach phrasal verbs.

Despite the tension between dictionary compilers who worked for commercial publishers (like Atkins) and dictionary buffs in academia, we learn in the conversation with Rundell how Atkins was a founding member of EURALEX, was elected to its first board (as secretary, naturally), and later became its president. She talks about the Hector Project, the BNC, WASPS, Word Sketches, the Sketch Engine, GDEX and DANTE – all of them milestones in lexicography. Take the Hector Project, for which she was invited to Palo Alto, California, together with Patrick Hanks and Rosamund Moon, where they were dazzled by having banks of six simultaneous screens but also quickly realized the Palo Alto engineers were not really interested in language nor the development of a proper language tool. Even stronger: “Every time it crashed, it was wonderful for them, because they knew what query had crashed it” – it, it later turned out (though they had no idea at the time), was AltaVista, an early Web search engine that was being developed. About the struggles to create the British National Corpus (BNC), we learn that Della Summers (from Longman) came up with the important concept of viewing a corpus as “a pre-competitive resource,” meaning “we could all use the same corpus and produce completely different dictionaries.” To date, this simple concept has still not been grasped, as may for instance be deduced from the latest survey on licensing lexicographic data (Reference Kosem, Nimb, Tiberius, Boelhouwer, Krek, Gavriilidou, Mitits and KiossesKosem et al. 2021b).

In order to fully grasp the genesis of the revolutionary COBUILD project, the reported conversations between Sue Atkins and her brother are revealing. Sue Atkins: “The real trouble about lexicography is that by the time the dictionary is in print, you’ve thrown away 90 percent of what you know about the words. And if you don’t throw that 90 percent away, no one would ever use the dictionary.” To which John Sinclair replied, “Well, I have a way to keep the 90 percent” and he would then explain how he first wanted to build a corpus.

Similarly, the reported conversations between Atkins and Fillmore give us an insight into how the American linguistic icon came to be involved. Fillmore: “I have to admit that you are not going to persuade me that I need a corpus. Because I know that in my head is what I know about language, and that is what I use when I am writing.” Atkins responds with “a little talk on RISK” and asks what the difference is between sentences like “There are too many risks there” and “There is too much risk involved.” Fillmore couldn’t tell and conceded: “I am really terribly sorry; I must now withdraw my comments.” Atkins and Fillmore then went on to work on “Frame Semantics Light,” which could be used by lexicographers.

What may come as a surprise to some readers is that Sue Atkins seems to have been particularly fond of her experiences in South Africa. About Nelson Mandela she says that “he felt particularly strongly that if you didn’t have a dictionary for your native language, you perpetually felt at a loss; you are identified as a second-class citizen.” She thus happily took up the invitation from Penny Silva to teach in South Africa but, tongue-in-cheek, claims: “I did all the wrong things! For instance, I’m only gonna give you half an hour on headwords, because of course headwords are so easy. And it turned out that the most difficult thing of all in many of the African languages was the headword. Which of the many, millions sometimes, forms of the word could you consider the headword?” When she later gave an assignment to compile an article for the word green, it came as a surprise that “many of the languages in the room had no word for green!” Clearly, this was a two-way learning exchange.

A last outstanding feature which needs to be mentioned as it is so characteristic of Sue Atkins is that, throughout the interview, she assigns vignettes to almost all the people she mentions, in order of appearance: Alain Duval, “a great friend and absolutely wonderful linguist”; John Sinclair, “a typical brother in every possible way”; Juri Apresjan, “a great lexical semanticist”; Krista Varantola, “a very honest translator”; Tony Cowie, “a brilliant lexicographer”; Antonio Zampolli, “the lifeblood of European computational linguistic research in those days”; Nicoletta Calzolari, “an extremely gifted linguist”; Beth Levin, “the most meticulous researcher I have ever met”; Igor Mel’čuk, “a brilliant academic lexicographer”; Chuck Fillmore, “the source of all wisdom for lexicographers”; Patrick Hanks, “a great lexicographer and a very-very gifted linguist, especially in lexical semantics; and a very good project manager”; Rosamund Moon, “a distinguished academic and lexicographer”; Jeremy Clear, “the computer whizz at COBUILD”; Nick Ostler, “an extraordinary linguist”; Uli Heid, “a brilliant organizer and a brilliant linguist; and very nice, very pleasant”; Penny Silva, “an absolutely organized, wonderful woman; and brilliant lexicographer”; Michael Rundell, “the most nice person I could live with for two weeks, and yet is competent and would do things really nice, and turn up on time, and know what he was doing”; Danie Prinsloo, “like living with a charged dynamo”; Gilles-Maurice de Schryver, “undoubtedly even more dynamically charged”; and Adam Kilgarriff, “a lateral thinker and completely brilliant.” Probably only a seasoned lexicographer like Atkins could have kept the pace, diligently filling in the vignette slot.

31.3 The Future of Dictionaries: Thinking Out of the Box

So what have we learned so far? Yes, stunningly innovative concepts and tools have been worked on over the past half century, and Sue Atkins was at the center of most of it. But perhaps, lexicographically speaking, Indo-European languages like English or French are simply “too easy” to warrant the development of truly innovative dictionaries, which in turn might explain the frustration some of us experience at the lack of true progress. It starts with the lemma. If it is non-problematic to lemmatize the words from each of the word classes (four morphological forms for verbs in English, three for adjectives, two for nouns, and for most other word classes just one), then, yes, a digital dictionary will simply be a calque of the paper dictionary when it comes to the macrostructure and the question of how to approach the lemmata. But consider a language in which one simply does not even know where to begin the lemmatization. Every attempt to produce a paper dictionary will look different, as every lexicographer will have their own approach, but this can easily be “solved” in a digital environment merely by not forcing any type of lemmatization. Users of such a dictionary simply search for orthographic words as they are pronounced or written, or as they are heard or read. A recent case study is provided by lexicographic work on Hupa, a Native American language of northwestern California (Reference SpenceSpence 2021). While it is laudable that an attempt was made to find the “logical” forms which should serve as lemmata, it seems so much more straightforward for polysynthetic languages like Hupa to employ the power of a computer to do the decomposing and recomposing for the digital dictionary user: simply allow users to input full orthographic forms and give explanations (or translations) for those forms. That would also mean that the dictionary actually has no lemmata and no real macrostructure, but so what? That is truly thinking out of the box.

After having looked into the macrostructure, traditional lexicography dictates that we then need to discuss the microstructure of future dictionaries, followed by possible mediostructures, megastructures, and so on. But this approach is ineffective, I am convinced, as every aspect of future dictionaries, as well as lexicography in general, will be radically different. To illustrate this, I summarize my thoughts using about fifty “oppositions” in which I contrast the past half century with my extrapolations for the future. These oppositions are grouped into five subsections and then recapped in a table after each one.

31.4 The Future of Lexicographers: Joining Forces with Big Data

During his keynote speech at the EURALEX 1998 congress, Gregory Reference 720Grefenstette, Fontenelle, Hiligsmann, Michiels, Moulin and TheissenGrefenstette (1998) famously asked, “Will there be lexicographers in the year 3000?” Inasmuch as lexicographers are the people who compile dictionaries, the fifty-odd “past ⇔ future oppositions” presented in the five subsections of Section 31.3 should lead to a clear answer because the question can be rephrased as whether or not there will still be dictionaries in the year 3000. Sadly, the answer is obvious: apart from legacy dictionaries, no doubt all of them digital or retro-digitized, no new dictionaries as we know them today will be found a thousand years hence or even a hundred. Unless, of course, both the concept of a dictionary and the job of a lexicographer are redefined. This may be compared with the word chauffeur for a person driving wealthy people around. Initially, a big part of the job of a chauffeur was actually both to stoke a steam engine and keep it running, with the word chauffeur derived from the French verb chauffer ‘to heat.’ Today’s electric cars only use power from the grid, no heat whatsoever is produced in the car, but chauffeur stuck. A decade ago, at the eLex 2011 conference, Michael Rundell, Erin McKean, Adam Kilgarriff, and others got ahead of themselves and openly wondered: “Will there still be dictionaries in 2020?” (see https://videolectures.net/elex2011_bled/, Round Table). The year 2020 came and went, and there are still (some) dictionaries, and there are still (some) lexicographers. That raises this question: Reconsidering the fifty-odd oppositions, is there anything common across them? And if so, can the commonalities be joined together and brought back to the current concept of a dictionary?

In Reference de Schryverde Schryver (2003, 146), I suggested viewing digital dictionaries as “collections of structured electronic data that can be accessed with multiple tools, enhanced with a wide range of functionalities, and used in various environments.” Laurence Urdang, the icon of early American dictionary making with computers, sincerely hoped that was not my definition of a digital dictionary (personal communication, email June 18, 2003), perhaps because an important feature is missing: a focus on the lexicon. But that was by design, as I have always felt uneasy with the artificial distinction made since Reference BloomfieldBloomfield (1933, 274): “The lexicon is really an appendix of the grammar, a list of basic irregularities.” But point taken – Urdang was right: what is common and what defines dictionaries and lexicographers is that they focus on lexical analysis, on meaning, and that they do so by studying and cataloging actual uses of words, collocates, phrases, and much larger chunks of text – these days as seen in unimaginably large quantities in corpora and thus with the help of computers to number-crunch the data and present humans with summaries. Corpora can be raw, but one obtains better results after lemmatization, part-of-speech tagging, and so on. In addition to not explicitly mentioning the “lexicon” nor “lexical analysis,” my earlier characterization of digital dictionaries did not even refer to “language,” whether in its oral or written form – that was by design. With my generic “electronic data” I wished to refer to the bits and bytes, the zeros and ones, of any type of communication, even of the non-verbal type, in any type of medium.

Surely, in future dictionaries one will be able to access the data via signing, smell or touch, with possibly only images, audio, or video in return. In such dictionaries, there is no lexis, no language. Yet information and thus meaning is still conveyed, and the actions still have to be decoded and thus defined. That is the task of a lexicographer, even a future one; that is the task of a dictionary, even a future one. Search the dictionary of the future merely by raising your middle finger toward it, to hear that your sign is used to express displeasure but also that it is rude to signal like that, or to have your dictionary automatically fill in its Unicode character (U+1F595) in the code you are writing. Place your dictionary next to the dish you are about to eat in a Mumbai restaurant and have it pick up the aromas to inform you about the spices and herbs it contains or to present you with pictures of the plants from which the spices and herbs are sourced (Reference HanksHanks 2012, 81). Or tap away a Russian folk tune on your dictionary, and hear and see how Beethoven recomposed it into his Razumovsky string quartets (Reference FerragutoFerraguto 2014; Reference KumarKumar 2020). That is the future of our dictionaries.

Future dictionaries, then, will have more in common with today’s search engines, e-commerce platforms, mobile apps, social networks, personal computers, and (micro)blogging sites, than with current dictionaries. But will future lexicographers also have more in common with today’s data scientists who already work at the companies developing such tools than with current lexicographers? If we are not attentive, if we continue to sleep at the wheel, our profession will indeed be gobbled up by them. So what can we offer them that they now miss? Answer: our craft – as in The Art and Craft of Lexicography (Reference LandauLandau 2001). We are still better than machines (and will likely always remain better) in taking computer-generated summaries on how language works and on how language is used, to derive meaning from that (Reference Hanks and CorréardHanks 2002; Reference Rundell and CorréardRundell 2002). Lexicographers will always better understand lexis, language use, and all types of meaning, and be better equipped to interpret and synthesize it, than people in any other profession, and certainly better than machines, no matter the sophistication of the machine learning or the artificial intelligence (AI) involved. (With the recent arrival of generative AI chatboxes like ChatGPT, I am inclined to want to revise this statement. I am now of the opinion that the future-future has arrived, also in lexicography. For the impact on dictionaries, see de Schryver 2023.)

With characteristic insight, Sue Reference Atkins, Kiefer, Kiss and PajzsAtkins (1992a, 519) once pointed out: “Lexicographers have to be born before they are made.” This led Edward Finegan to observe: “Her statement haunted me a bit as discussion of lexicographical automation may suggest that lexicographers don’t even need to be made – don’t need to exist – so long as the AI folks can finesse their algorithmic magic” (personal communication, email September 10, 2021). I hope to have shown that the Big Data technicians will never be able to supplant us, as long as we continue to convince their employers that they need our skill set, our know-how, and our scientific research.

31.5 Epilogue: A Fact-Based Extrapolation of the Future

Rather than jump right into “the future of dictionaries” to describe that future in medias res (for, frankly, how can one jump into the future when that future has yet to occur?), I have opted not to hypothesize about the future of dictionaries at all but to base all my claims on facts that can be observed today, and to extrapolate from them.Footnote ² What this approach demonstrated about dictionary making is how a real person, who was alive when some readers of this book were (for some time) also alive, changed the course of lexicography toward the future. So lexicographers make the future. By opening and closing with Sue Atkins, this chapter not only fleshes out a recent super-lexicographer but also articulates her network: her epithets for her colleagues show how linguists, lexicographers, and computer scientists work together in developing new approaches to the lexicographer’s work, with the aim of producing radically different types of reference works. Ironically, the core element of Sue Atkins’ vision of the “virtual dictionary,” uttered a quarter century ago, namely that the information provided will exist only at the time of consultation, is now everywhere (think for instance “map,” and see how it forms around you, on the screen in the palm of your hand, with you at the center of it all at all times, even as you move through space and time, as it indeed changes with time) – it is ubiquitous, yet it passed lexicography by. Alphabet (what’s in a name?) did it for general and factual information, Amazon for shopping, Apple for mobile assistants, Meta for social networking, Microsoft for personal computing, and X (formerly Twitter) for micro-blogging – truly, data is the new oil (The Economist 2017) – but Lexicography is not yet there. Rather than have the Big Data companies attempt to do our work, we should convince them that they will be better off by joining forces with Lexicography. Lexicographers have something unique to offer – it is in our DNA; only we know how to get to the bottom of how language works, only we have the craft to derive meaning from usage, and only we are able to synthesize this – beautifully – for both human and machine consumption, in any number of ways.

In future, our lexical analyses will not be wrapped in dictionaries; they will disappear from view to become data among ever more data, in a Linked Data network. But if done right, they could become the “crown data” for which the Big Data companies will be willing to pay good money, as with our crown data their tools and products will both be and sell better than they do today. Lexicographers should therefore continue doing what we already do: trust data, lots of it; have machines do the heavy lifting in pre–processing; then come in for the invaluable finishing conceptual and lexical touches. Given that we have now given up on the idea of the need to work face-to-face, future all-virtual lexicographic teams will become ever more international, will incorporate ever more varied but complementary skills, will work on ever more languages and language varieties, and will store their data and analyses in and license them from the cloud – all so that lexicographers will be able to serve users and machines with their ever-growing lexical needs.

Steve Jobs, co-pioneer of the personal-computer revolution and co-creator of a range of i-tools, is known to have said: “A lot of times, people don’t know what they want until you show it to them” (Reference RatcliffeRatcliffe 2018). That should also be the guiding principle of future lexicographers (and metalexicographers): stop asking dictionary users what they want; just make the best possible dictionary tools; and study how they are used. Then share your research results online in continuously growing repositories, approached and searched with the help of increasingly smarter access and summarizing software.

There is, then, a clear path toward future dictionaries and lexicographers. If we seize the moment, both may thrive going forward.