CityInnocent misperception: the linguistics of Starbucks name fails
Are #StarbucksFail errors innocent? In order to test this, we can see if the changes between the target and error names look like common sound changes in the world’s languages.
Linguistics occupies an interesting position among the sciences. It has its fair share of brandy-swilling and argyle sweaters, but also has the responsibility to be inclusive of new trends in language, wherever they may occur.
I stumbled on a treasure trove of data in Starbucks, the kind of data that eludes my colleagues in white coats. As linguistic work encroaches on increasingly casual domains, its content only becomes more accessible to people outside the ivory tower.
One of the biggest questions linguists (and in particular, phonologists) try to answer is what the smallest unit of speech sounds are. The most accepted view is Distinctive Feature Theory, that boils sounds down to properties (“features”) such as [voice] for sounds made with vibrating vocal cords or [labial] for sounds that make use of the lips. These combine to form actual phones like [b], a sound that has both of these properties. Such a broad theory is naturally difficult to test, and linguists have had to turn to creative modes of experimentation to gain any insight.
Ben Macaulay is a PhD student at the CUNY Graduate Center specializing in tonal phonology and morphosyntax
One study, the Fromkin Speech Error Database, catalogued all the slips-of-the-tongue that happened at a series of conferences to see if these speech errors differed from their intended utterances by these features. For example, a speech error whose target is “cat” might surface as “gat,” “cad,” “cot” etc. What Fromkin found, however, was that these errors in production were more likely to do things like splice words together (“brun” for “bread bun”) or be spoonerisms like “pimming swools.”
I decided to look at speech errors from the other side—perceptual errors as opposed to the production errors in Fromkin’s database. The idea dawned on me one day in Starbucks, when my friend Wataru and I received drinks labelled for people named “Zen” and “Retardo.” Using hashtags like #starbucksfail on Twitter and Tumblr, I was able to compile a corpus of the wrong names people get at Starbucks and how they compare to the submitters’ actual names. Because it’s impossible to plunk people down in a laboratory setting and get speech errors out of them, the large amount of self-reported data surrounding Starbucks errors is a unique opportunity to generalize over a large body of errors. The corpus amounted to 115 points of data after filtering out spelling errors (failing to select the correct vanity spelling of “Cailyn”), obvious drink thieves (“Why does my drink say ‘Frank’? My name is Hannah!”) and unpronounceable results (e.g. “Jhamie” and “Lieaia”).
Using hashtags like #starbucksfail, I was able to compile a corpus of the wrong names people get at Starbucks and how they compare to the submitters’ actual names, amounting to 115 points of data after filtering out spelling errors, obvious drink thieves and unpronounceable results.
Because the “before” and “after” states of sound patterns that undergo change are generally similar perceptually, these changes are truly accidental.
In order to test this, we can see if the changes between the target and error names look like common sound changes in the world’s languages
Very few of the data had changes that were not along feature lines or mirrored natural sound changes. It’s safe to say that Starbucks is not “fucking with you.”
JareD → JeroL
A common type of sound change —
lenition (“weakening” of a sound)
Popular culture has something to say about why these errors happen. According to a popular piece of YouTube smarm, Starbucks workers are just “fucking with [us],” that is, they are intentionally picking out wrong names that are just-wrong-enough. And to their credit, many phonological frameworks (such as the most popular these days, Optimality Theory), work this way too. They assume that changes in sound patterns are teleological, or “goal-oriented” and that whether it be changes to entire languages, loanwords, or simply speech errors, these changes make sound patters better in some way (or in their words, “less marked”).
The opposing view, both in Starbucks and language at large (Evolutionary Phonology, for example), is that these changes are truly accidental. The argument here is based in acoustic properties of sounds: because the “before” and “after” states of sound patterns that undergo change are generally similar perceptually, these changes happen because of innocent misperception, and not some conscious goal.
With this in mind, we return to our original question: are Starbucks errors innocent? In order to test this, we can see if the changes between the target and error names look like common sound changes in the world’s languages. I looked at two things: whether the changed sound is perceptually-similar to the expected sound, and whether the change occurred in an environment where it was expected. Let’s look at the results (on the right).
A common type of sound change is fortition (“strengthening” of a sound) and lenition (“weakening” of a sound). Fortition often happens at the beginning of a word, while lenition is common in consonants between vowels. Fortition occurred 12 times word-initially, 5 times intervocalically and once word-finally. Lenition occurred 6 times initially, 3 times intervocalically and once finally.
Fortition: Vlad → Blad, Anna → Hannah
Lenition: Kayla → Gayla, Cody → Zoey
Fortition: River → Riber
Lenition: Vlada → Lana
Fortition: Ellis → Alex
Lenition: Jared → Jerol
GaiL → CatO
Syllable added due to confusion between
final unstressed syllables and falling tones
While changes between vowels were more likely to be lenition, the probabilities of fortition/lenition in these environments were a lot closer than expected. It seems that there were “problem spots” where people had trouble hearing perceptual cues, but whether they fortited to overcompensate or lenited was up in the air. This kind of overcompensation is likely, as errors were more than twice as likely to increase voice onset time (e.g. d → t) than to decrease it (t → d).
Another feature that changed often was place of articulation, either major (which organ is used, e.g. m → n shifts from the lips to the tongue) or minor (what that organ does; e.g. s → th shifts the tongue from the roof of the mouth to the teeth).
Other changes involved syllable structure, in various ways. Syllables were added or deleted, clusters (groups of consonants) were simplified to single consonants and there was confusion between final unstressed syllables and falling tones (a sound change common in dialects of German and Dutch).
In fact, very few of the data had changes that were not along feature lines or mirrored natural sound changes. There were a couple who heard part of the name and filled in the rest with some other word (Valeria → Vanilia). One might ask whether all of these changes are people filling in more common names based on hearing partially, however based on the rankings at namestatistics.org, errors that were also real names were split between raising (Ryan → Bryan) and decreasing in frequency (Jessie → Chelsea).
It’s safe to say that Starbucks is not “fucking with you.” In fact, as an observable process of sound change (in the form of errors), the identity crisis on the side of your Pumpkin Spice Latte is another nail in the coffin of goal-oriented phonology. In a way, every word we speak winds up on the side of someone’s Starbucks cup. Over time, our baristas make so many similar mistakes that our language takes on a new form. It’s not their fault, this is just how languages change, and without this process the world’s languages would be radically different from how we know them now.
Tom → Ton
Ben → Dan
Addy → Abbey
Kelsey → Kelthy
Joya → Troya
EPENTHESIS (SYLLABLE INSERTION) (4)
Kathryn → Katheran
Emeka → Amneka
SYNCOPE (SYLLABLE DELETION) (8):
Felicia → Alicia
Augustus → Custer
CLUSTER REDUCTION (3):
Brianna → Rihanna
UNSTRESSED SYLLABLE ↔ FALLING TONE (3):
Lyla → Lyle
Gail → Cato