by Sloane Goldberg by Sloane Goldberg
art by Sophie Kamdar
On June 12, 2023, to the delight of procrastinators everywhere, The New York Times debuted a new game, Connections. The goal is simple: the game gives you sixteen words that can be divided into four categories. The first, yellow, is the easiest; green is a little harder; blue is tricky; purple sometimes feels downright impossible [1].
Most of the time, the game makes sense. What category do Hum, Scat, Sing, and Whistle belong to? According to Connections, they’re all terms for carry a tune. Or, after a bit of thinking, a player could realize that Channel, Line, Main, and Pipe are all words for water and gas conduits. But what about Lucy, New York, Rock and Roll, and You, or Face, Hurricane, Potato, and Needle?
It quickly became clear that Connections had a major trip-up: one of the categories, usually purple, made no sense. Annoyed users began to complain about categories, such as things you can set or distinctive number of arms (or lack thereof ), or my least favorite, hail size comparisons, which featured Grapefruit, Baseball, Marble, and Pea. If anyone has ever experienced grapefruit or baseball-sized hail, let me know.
Yet the game also has a second, more insidious hurdle: when you open Connections on The New York Times site, the game loads as a 4x4 grid of squares, each containing one word. A new user might think these tiles are randomly arranged, swirled around on the floor like a children’s matching game. But, dear Connections player, this is just another one of The New York Times’ tricks.
To understand how Connections fools us so easily, we must first take a brief detour and analyze our attention. When we click or tap on the Connections game and its little squares float onto the page, our visual field tries to comprehend the salient information, or the information that is relevant to our understanding, from the scene as fast as possible. How do we do that?
One study created a model of attention surveillance by tracking participants’ eye movements as they searched for items in a scene [2]. First, researchers gave a word cue (like trash can, drinking glass, or painting), and then gave participants an image with that item in the scene. Participants were instructed to mark how many of the items they saw in the scene within 10 seconds. Using eye-tracking software, researchers quantitatively measured the way their eyes moved around the screen as a measure of where their attention was localized [2]. One of the study’s conclusions was a significant interaction between key areas of the room and target-related surfaces; in other words, while participants searched for a drinking glass, their eyes scanned tables, counters, or desks where a glass might be. This is important because it indicates that people used preconceived notions about items (e.g., garbage cans are on the floor, but drinking glasses are on the counter) to direct their search.
Let’s pause this study for a moment and look at one more. A 2014 study by Potter and colleagues looked at how long it took participants to identify what they were looking at. Researchers used a technique called rapid serial visual presentation (RSVP), where participants looked at a series of images, rapidly shown one after another to examine visual attention and perception. In this task, they were given between six and twelve images, each presented between 13 to 80 milliseconds. Researchers then asked viewers to detect a specific photo by giving them a name like swan or bear catching fish before or after viewing the images. Participants were also asked to identify when two images might fit one title. Their study showed that participants could conceptually understand the image in as little as 13 milliseconds, even when they weren’t given the title of the image until after it was shown [3]. In 13 milliseconds, and when the target image was sandwiched between other irrelevant photos, the accuracy of a participant’s understanding of an image was still above random chance.
Knitting these two studies together, we see both how quickly and how accurately people can understand images. Within a hundredth of a second, humans can visually process a scene [3]. Once someone understands what they are looking at, they can search the image for relevant information. These studies explain some of the basic elements of Connections. First, when you look at the screen, the game limits your search for appropriate words to the set of 16 words given; you’re not looking at the search bar or the copyright footer for game words.
However, we can go a little deeper. Without consciously looking for something, it could take someone as little as one-hundredth of a second after the screen loads to understand the first word of the grid. Once someone sees the first word, their eyes may start to flit around the screen looking for where its matches might be. In the same way that participants in Potter et al.’s study who saw the word “trash can” and understood both the meaning and other things it may be associated with (e.g. the floor, garbage), a Connections player reads the scene and begins to search for relevant information before they may have even consciously processed what they’re looking at.
Ok, back to The New York Times game. Connections loves to use words with multiple meanings and words that fit into multiple categories. Take, for example, the word bun. It could mean a bread bun, like with a burger, or be part of a saying like bun in the oven, or a hairdo, or maybe even be a word for butt. The second I spotted the word bun in the corner of the game, I would begin to scan for similar words, viewing everything through how it might relate to bun. The next word that catches my attention immediately: salad. Word association: bun… salad–side dishes! What other side dishes are there? …Rumble? …Pony?
I’m sure the way my thought process devolved is familiar to anyone who has played Connections. But what went wrong? How was my millions-of-years-evolved brain stumped by three words in a word puzzle? I could see the scene, and I could infer related words, but that was where my success dropped rapidly.
Therein lies the trap: Connections relies on you to make initial guesses that are wrong. Instead of immediately correcting a mistake and starting over, the game tricks you into thinking you’ve found the correct solution, sinking you deeper into trying to prove an ultimately incorrect answer.
Looking at the puzzle in which I got sidetracked by lunch words, I can find a simple explanation for why after bun, I looked to salad: they’re right next to each other on the grid. Bun is also bordered by root and sift, but because those have no obvious connection, my brain rejected them in favor of an answer that made associative sense, in the same way study participants focused their search for a drinking glass on tables, not the floor.
The best example of this is from the February 19, 2024 Connections game. The first two rows read: Hinge, Spoil, Drill, Match / Lock, Tinder, Handle, Bumble. Because of the order in which the words were presented, most readers pulled out Hinge, Spoil, Drill, Match / Lock, Tinder, Handle, Bumble. Clever, right? Dating apps! Not so fast.
Imagine if the words were presented like this: Lock, Handle, Hinge, Bumble / Spoil, Drill, Match, Tinder. With the correct themes grouped next to each other, the right connections pop out: Lock, Handle, Hinge, Bumble / Spoil, Drill, Match, Tinder. Through the context of Lock and Handle, Hinge isn’t a dating app anymore. Rather, it’s a door
part. Without the prior contextual framing of Hinge, the terms Match and Tinder become firestarters.
The concept of contextual framework is a form of the priming effect, a powerful psychological tool that subconsciously changes the way we interpret information. For example, if you’ve spent the last hour shopping for shoes online, you might notice the shoes of the next person you meet before any other aspect of their outfit [4]. Psychologists use the priming effect as a way to understand how previously viewed information impacts the way participants understand the information they’re currently viewing.
Specifically, Connections relies on the semantic priming effect, a version of the priming effect that changes how one processes a word based on the word read before [5]. In fact, one study from a few years before Connections was invented looked at how the presentation of words impacted how quickly other strings of letters were recognized either as words, or as meaningless jumbles of letters when the two words were related. Researchers used an ocular lexical decision task, where they told participants to indicate whether they thought a string of letters was a word or not by moving their eyes across the screen. If the string of letters, like friend was a word, participants glanced toward the “word” category. If the string was not, like friendically, they glanced towards the “not a word” category. When the two words next to each other were related, like mischief and trouble, (or like in our previous example, bun and salad) participants more quickly recognized that the second string of letters was a word and moved on to the next word– even though they didn’t consciously acknowledge that the two words were related. This effect was not seen with two words in a row that were not related, like fog and train [5].
So, when we see words that seem similar, we read them faster, indicating that we are registering their meaning internally. That’s why the order matters, and why the first line of a Connections game from February 7, 2024, read Sponge Bob Square Pants (and, dear reader, by this point you should not be shocked to hear that those words all fell into different categories). Even if you internally register that those words obviously don’t share a category, it’s so hard to shift gears and look at other answers once you’ve read the question that way.
But why are we so susceptible to priming? And why is it so hard to change the way we read something? It feels like such a simple way to fool our complex and diverse brains. Well, it turns out there’s one very important evolutionary reason for why priming works, and why it exists: when given limited information, priming allows us to make faster and more accurate decisions [6].
The answer lies in the two main ways we look at information: top-down, and bottom-up. Top-down, or goal-directed processing, is slower and more cognitively taxing, as it is thinking that originates in higher-order regions of the brain, like the cortex, which drive decision-making, goal-oriented behavior, and actions [6]. Examples of active top-down thinking include thoughts like, “If I want to pass my exam, I should start studying now,” or, “What should I pack for a snack?” These thoughts rely on past experiences, future goals, and available information. Bottom-up processing works in the opposite way and includes observing things around you, like your open computer or an apple on the counter, and forming thoughts around them. Passive bottom-up processing is faster and relies on a certain level of uncertainty, as the world is full of competing sensory inputs and scenes with no apparent connection. Without bottom-up processing, it would be impossible to pick just one thing to focus on. Researchers hypothesize that the priming effect exists to reduce the uncertainty in bottom-up processing; it’s easier to separate competing sounds, smells, and colors if you’re specifically trying to match observations to what you’ve seen before [6].
This bottom-up processing drives a vast portion of our unconscious thinking. One study showed that 75% of people choosing food from a buffet picked the items that were offered to them first; people in lines for unhealthy foods shown first ended up choosing more food overall, and more unhealthy food [7]. Another study on politics gave participants real and fabricated news articles, then asked them if they thought what they had read was “real news” or the famous, “fake news.” Participants in the group that were primed with information about the prevalence of “fake news” struggled to identify “real news” articles, indicating that what they had heard before was overriding their ability to independently determine if something was true or not [8]. We are inherently at the whim of what we have seen before.
Yet, understanding isn’t a lost cause. One final study showed that while unexpecting participants were subject to the priming effect, participants who were prepared fared better. Researchers used a flanker task, where a letter of interest is bordered by two competing symbols on either side, like F O F. In the past, researchers believed that the introduction of the competing characters always led to slower response times. However, researchers found that using symbols that looked like a mix between 5/S or 0/O introduced ambiguity about whether participants were looking at letters or numbers. So, in trials where participants were previously given letter only trials, the ambiguous 5/S read as an S, and participants picked out the number between them (i.e. S 9 S). On the other hand, when participants were previously given trials with numbers between the letters, like F 4 F, the 5/S was interpreted as a number (i.e. 5 9 5), and they struggled to differentiate. This study provides evidence that differentiation is not a passive process, but an active one, impacted by contextual clues [9].
So where does that leave us with the enigma that is The New York Times Connections? Well, as the last study indicates, expectation is half the battle. If you expect the first line to not show you the answer, that will help pick out incongruent options. Maybe, per the first study, you could start the game by looking at a different part of the screen each time so you’re forced to move your eyes around in a new way [2].
Or, take the laughably simple approach I learned from my younger brother. Load the game, then immediately hit the “shuffle” button at the bottom of the screen. Like waving a magic wand, the carefully curated priming disappears, leaving you to match categories, sight unseen.
REFERENCES:
[1] Liu, W. (2023, June 26). How Our New Game, Connections, Is Put Together. The New York Times. Retrieved from https://www.nytimes.com/2023/06/26/crosswords/new-game-connections.html
[2] Peacock, C., Cronin, D., Hayes, T., & Henderson, J. (2021). Meaning and expected surfaces combine to guide attention during visual search in scenes. Journal of vision, 21, 1. https://doi.org/10.1167/jov.21.11.1
[3] Potter, M. C., Wyble, B., Hagmann, C. E., & McCourt, E. S. (2014). Detecting meaning in RSVP at 13 ms per picture. Attention, Perception & Psychophysics, 76(2), 270–279. https://doi.org/10.3758/s13414-013-0605-z
[4] Stevens, W. D., Wig, G. S., & Schacter, D. L. (2008). 2.33 - Implicit Memory and Priming. In J. H. Byrne (Ed.), Learning and Memory: A Comprehensive Reference (pp. 623–644). Oxford: Academic Press. https://doi.org/10.1016/B978-012370509-9.00150-9
[5] Hoedemaker, R. S., & Gordon, P. C. (2017). The Onset and Time Course of Semantic Priming during Rapid Recognition of Visual Words. Journal of experimental psychology. Human perception and performance, 43(5), 881. https://doi.org/10.1037/xhp0000377
[6] Race, E. A., Shanker, S., & Wagner, A. D. (2009). Neural Priming in Human Frontal Cortex: Multiple Forms of Learning Reduce Demands on the Prefrontal Executive System. Journal of cognitive neuroscience, 21(9), 1766. https://doi.org/10.1162/jocn.2009.21132
[7] Wansink, B., & Hanks, A. S. (2013). Slim by Design: Serving Healthy Foods First in Buffet Lines Improves Overall Meal Selection. PLOS ONE, 8(10), e77055. https://doi.org/10.1371/journal.pone.0077055
[8] Van Duyn, E., & Collier, J. (2019). Priming and Fake News: The Effects of Elite Discourse on Evaluations of News Media. Mass Communication and Society, 22(1), 29–48. https://doi.org/10.1080/15205436.2018.1511807
[9] Avital-Cohen, R., & Tsal, Y. (2016). Top-Down Processes Override Bottom-Up Interference in the Flanker Task. Psychological Science, 27(5), 651–658. https://doi.org/10.1177/0956797616631737
Comments