From Chris Waigl on Facebook yesterday, coping with the day’s Spelling Bee game on the web, in which she was told that her candidate UNVOICING was not a word — well, not a word acceptable in the game. Her hedged response:
UNVOICING is a word. (Well, maybe.)
(CW is, among other things, a linguist, and linguists often have complaints about what Spelling Bee is willing to accept as a word of English.)
I’ll expand on CW’s comment, and that will take us to a surprising place (AI chatbots and their discontents). But first, some background on the NYT Spelling Bee.
Spelling Bee. From Wikipedia:
The New York Times Spelling Bee, or simply the Spelling Bee, is a word game distributed in print and electronic format by The New York Times as part of The New York Times Games. Created by Frank Longo, the game debuted in a weekly print format in 2014. A digital daily version with an altered scoring system launched on May 9, 2018.
The game presents players with a hexagonal grid of 7 letters arrayed in a honeycomb structure. The player scores points by using the letters to form words consisting of four or more letters. However, any words proposed by the player must include the letter at the center of the honeycomb. Each letter can be used more than once.
A partially completed version of the August 8, 2021 puzzle. The sole pangram of this puzzle was “inflect”.Scoring points leads to progressively higher praise for the player’s effort, such as “Solid”, “Amazing”, and “Genius”. Each puzzle is guaranteed to have at least one pangram, which awards the player the largest number of points when found. If the player finds all of the possible words in a given puzzle, they achieve the title of “Queen Bee”.
The form UNVOICING. What I said on Facebook, somewhat edited:
A number of dictionaries have the verb UNVOICE as a synonym for the more usual DEVOICE; for this verb we’d get the form UNVOICING. But the technical usage for the change-of-state verb is DEVOICE ‘make voiceless, remove voicing from’ (with forms DEVOICING and DEVOICED), not UNVOICE (even though the prefix UN- is often used for reversative verbs, like UNTANGLE).
The technical usage, however, is a minefield, and if you’ve ever taught phonetics — as I have, as part of introductory linguistics — you will appreciate what a minefield this is for students: in technical talk, the opposite of the stative adjective VOICED ‘articulated with voicing, with vibration of the vocal cords’) is not, as a reasonable person might have expected, UNVOICED, but VOICELESS. (In non-technical contexts, UNVOICED ‘not voiced’ is fine: They had unvoiced objections to the proposal, referring to objections they hadn’t voiced, hadn’t given voice to, hadn’t spoken).
The larger point is that technical terms are largely fixed, and synonym substitution will not do. (There are sometimes alternatives for technical terms, and sometimes the alternatives are (near-)synonyms, but such situations usually just tend to generate disputes within the relevant technical community as to which alternative is the “right” one.)
In any case, they are largely fixed, and you can’t mess with them. They are like proper names: the actor’s name is William Hurt, and William Pain won’t do as an alternative; the musician’s name is Johnny Cash, and Johnny Money won’t do as an alternative; conversely, the psychologist and sexologist is John Money, and John Cash won’t do as an alternative. They are, in fact, like a large range of fixed expressions: the common name for the flowering shrub Ceonothus is California lilac, and Golden State lilac won’t do as an alternative; the idiom is eat crow, and eat raven won’t do as an alternative (though some idioms permit a certain amount of playful variation); and so on.
But not being allowed to mess with technical terms is a hard lesson for our students to accept, since they have been drilled for years by teachers that they shouldn’t over-use vocabulary in your writing, shouldn’t just keep repeating some word; instead, they should vary their wording to make it more interesting. This advice can lead to thesaurusizing, an overkill that deserves deprecation; but in the world of technical vocabulary it is absolutely contravened: in this context, there’s one right word, and you have to use it every time.
But now a twist: in many domains, technical vocabulary is almost entirely distinct from everyday vocabulary. The taxonomic labels of biology, for instance, are almost all scientific Latin creations that are clearly different from everyday-language vocabulary (though a few, like gladiolus, have been adopted as common names). Much other technical vocabulary uses the morphological resources of the language, but with learnèd roots, so that the terms are unmistakably not everyday vocabulary; much phonetic terminology is like this, for example:
coronal, obstruent, approximant, fricative, affricate, alveolar, implosive, uvular
But then there are ordinary words pressed into service as technical terms — technicalized, as I’ve put it — as when fly, bug, nut, and berry are adopted as terms in biology, with referents quite different from (though distantly related to) the ordinary-language terms, so that, for example, watermelons are technically berries. I appreciate the instincts that lead scientists to technicalize, but I’m on record as deprecating the practice, because of the confusions it so easily leads people into. Quite a few phonetic terms — trill, click, central, rounded, etc. — are (metaphorical) technicalizations, and they, predictably, lure people into thinking that (near-)synonyms might serve just as well (and would be available if you didn’t want to re-use the term): that twitter or chirp would do for trill, snap or pop for click, middle or medial for central, and round or oval for rounded.
As it happens, so many people have taken the verb unvoice as a synonym for the technical term devoice that many dictionaries just list it as such, and that then gives us the forms unvoicing for devoicing; and unvoiced for devoiced and voiceless. And, frankly, after all these years, I’m beginning not to care myself; what the hell, let the phonetic flowers of unvoicing and unvoiced bloom.
The fixity of fixed expressions. Technical terms among them. Not an easy thing to learn, which expressions are fixed, in which contexts, and to what degree. It’s a task for language-acquiring children, it’s a task for second-language learners, it’s a task for everybody when we move into new contexts and communities of practice, it’s a task for AI chatbots honing their abilities in mimicking human practices.
On the last of these: AI systems are not well-designed to learn fixity; they are agents of approximation. I’m not sure how they achieve some kinds of fixity (for proper names, in particular), other than by stipulation by their designers. So they sometimes run off the rails with other classes of fixed expressions.
An anecdote from my blogging life. I am inundated, every day, with various kinds of offers from entities that want to post guest material on this blog. These approaches to me used to be absurdly crude, employing as a hook some topic mentioned in one of my postings and offering to supply more content on that topic; it was clear that no one had actually read the posting in question.
Then came the new breed of bots, which paraphrased the content of my posting (usually pretty adroitly) — but regularly also paraphrased the posting’s title, thus giving the game away.
I imagine someone will find a way to stipulate that titles are inviolable, fixed, not amenable to approximation. In another cycle of learning.