What's in a Name? [News & Reviews]
How abstract representation can shape the brain and behavior
Welcome (back) to Brain Blast, your newsletter connection to the worlds of brain science, cognition, neuroscience, and psychology. This week we’ll explore how abstract representation – naming things, associating them, reasoning about them – fundamentally underlies how we understand and interact with the world. Today’s segments highlight how abstract representation has taken the spotlight in the news and in a recent academic publication. On Friday, you can look forward to a deep dive on abstract representation and how our brain’s language center uses it to give our minds vast creativity and flexibility – while also constraining us in subtle and interesting ways.
Click on the headings below to jump straight to a section!
In The News: Fine, I’ll write about the elephants
Review Corner: “Affect labeling” as a potential new PTSD treatment
In The News
Fine, I’ll write about the elephants
On June 10, 2024, researchers from Colorado State University announced the publication of their new study on communications between African elephants. This was reported on by most major news outlets, each with very unique and original titles such as “Elephants Have Names For Each Other Like People Do, New Study Shows”, “Every Elephant Has Its Own Name, Study Suggests”, and “Elephants Call Each Other By Name, Study Finds”. You might be able to guess that the bodies of these articles showed similar levels of individual reporter insight on the story, so pick your favorite as the main source.
To briefly summarize the main takeaways, the study authors have purportedly found that African elephants have, and use, “names” for each other. Quoting the original report by CSU, they used “machine learning” to “confirm that elephant calls contained a name-like component”, which they tested in real life by playing recordings of the calls back to the elephants and comparing their reactions to calls that referred to them versus to other elephants. The authors quoted in these articles, Dr. Michael Pardo (first author and postdoc researcher) and Dr. George Wittemyer (senior/last author and professor), contrast their findings to those from other non-human animals like dolphins, whose calls rely on mimicry of signature sounds directly made by their intended referrant. Imagine if, rather than your parents simply giving you a name when you were born, everyone had to wait around for you to make your first scream-cry noise, and then they shouted that scream-cry back at you every time they wanted to get your attention for the rest of your life; that’s dolphins.
At first, I wrote this off as a cutesy oversimplification of the probably nuanced and overstated findings of yet another overhyped Nature paper. But after not finding anything else particularly intriguing in the brain science news cycle this week, I figured I’d check it out in more depth. Like a good little scientist, I immediately went to find the original research report. After being blocked by the Nature paywall (booo publishing industrial complex booo), I quickly found the BioRXiv preprint1 and set out to better understand what the authors did and found. And it turns out, much like last week, that the news didn’t even pick up on the coolest parts!
First of all, the sheer scale of this study is positively elephantine (I do not apologize). The data come from as far back as 1986 in some cases and covers two separate geographical areas in Kenya. While the article doesn’t and probably can’t specify the exact number of elephants recorded, just the recordings actually analyzed in the study (and I’m sure it’s a fraction of the total collected over time) cover 114 unique “speakers” and 119 “listeners”. Moreover, the study team conducted the absolutely incredible task of learning the elephants by the shapes of their ears and determining their relative positions in their family’s social structure, then painstakingly tracking the specific locations of the elephants speaking and listening in relation to all the other elephants in the family at that moment every time a recording was taken. And that’s just collecting the raw data.
My own research efforts over the years weep at their inadequacy.
A quick background on the data and some terminology before we talk about what they actually found. One of the common ways elephants call to one another is through deep, complex sounds called “rumbles”. They make these rumbles in two key contexts for our purposes here: “contact rumbles” are used to call to an elephant that is out of eyesight of the main family unit or is far away, while “greeting rumbles” as the name may imply are done face to face when elephants reunite with one another. Notably, both of these are scenarios in which one might be compelled to use someone’s name. For each rumble, the authors identified a “caller” – the speaker – and a “receiver” – the listener. So, for each rumble, we would expect the “name” of the receiver to be part of the signal, if the caller indeed said it at all.
Now let’s talk about their key findings. What do they mean “elephants have names”? How do they know that? As the news outlets vaguely indicated, they used machine learning. Specifically, they found that they could sort out who was the receiver mathematically in 20.3% of the real calls relative to just ~8.3% in their null model. Being able to figure out the receiver from the signals themselves suggests that there is a consistency in the signals directed to a given receiver, much like your name remains the same in any sentence your friend might use it in. While 20.3% seems low, it’s actually a great find; this result included all of the recordings, since there’s no way for us to know if the caller actually referred to the receiver’s “name” at all. How often do you actually use your friends’ names when you say hi to them versus just “hiya, how’re you?” or the simple classic “heeeeeeey!” as you walk in the door?
Of course, the authors didn’t just look at the math and call it a day – they tested their conclusions by playing recordings back to the elephants and seeing how they reacted. They found that elephants reacted to calls with themselves as the intended recipient by responding verbally more often and more quickly and by approaching the source of the sound more quickly than if another elephant was the intended recipient of the call. If your friends shout your name from across the room, you’re more likely to head that way than if they shouted “Hey, Freddy!” (unless, of course, your name is Freddy; then imagine they shouted “Gertrude” to get the full effect).
So why should we care that elephants might have names? Names are a form of “abstract representation”, which is a way to mentally represent an object, concept, or what have you without being directly based on its physical characteristics. Your name represents you, and when your friends think of your name they think of you, even though they may not see you or even need to imagine you to do it. Likewise, I can tell you I have a friend named “Kaitlin” with red hair and a tattoo of a cat, and you can create a rough representation of this person in your mind, find them at a party, and creep them out by greeting them by name even though you have never met them before. Dolphins can’t do that until Kaitlin the Dolphin gets weirded out by them standing awkwardly in silence near her at the party and scream-cries at them trying to get them to go away. If elephants have abstract names, then perhaps they have more party decorum.
More seriously, the possibility that elephants have names suggests that they may more generally have developed a form of abstract representation. For humans, abstract representation is driven by our development of language; we can create words for anything, and we can use grammar and syntax to represent dynamic relationships between those words. Whether elephants’ use of names implies that they also have a greater language-based system is not yet known, but it’s not out of the question. Certainly if they do have language, it is wildly different than our own; there’s no simple repeated portion of the signal intended for a given receiver, like you could just insert someone’s name into a sentence. Instead, elephant rumbles carry multiplexed information, as if someone said “Hello” and your name simultaneously. Alternatively, they may not actually have language in any grammatical way, but may still have some other form of abstract representation that we can barely hope to understand, like Arrival without the aliens. Or maybe they just have names and that’s that; who knows.
Whatever the extent of elephants’ capacity for abstract representation, they seem to have developed it evolutionarily separately from our development of language, perhaps as a result of the pressures of managing a family unit that is often widely spread across a large geographical distance. In any case, the work these researchers have done to uncover this new discovery is truly impressive (remember, some data is from 1986), and their findings have some seriously cool implications for our understanding of elephant behavior and intelligence.
Review Corner
“Affect labeling” as a potential new PTSD treatment
Today's article, "Affect Labeling: A Promising New Neuroscience-Based Approach to Treating Combat-Related PTSD Veterans" was published open-source in the journal Frontiers in Psychology on May 7, 2024. It is a primary research report by Lisa J. Burklund, Carolyn D. Davies, Andrea Niles, Jared B. Torre, Lily Brown, Meghan Vinograd, Matthew D. Lieberman, and Michelle G. Craske, in affiliation with the University of California, Los Angeles, and NeuroGen Technologies, Inc2.
Today’s article reports a clinical study investigating a potential new therapy tool for combat-related PTSD in veterans. The authors conducted their study in 20 patients with diagnosed PTSD and 20 trauma-exposed healthy control subjects3. The base design for the patient side of the study was a pre-post design with six intervening “training” sessions over three weeks. That means that each patient came in for a baseline testing session, then completed six separate sessions with the new therapy tool (we’ll get into that), and then came back one more time for a final test and debrief session. The healthy control subjects just did the baseline testing at the beginning; this is what’s often referred to as a “no-contact” or “passive” control. During the baseline session, as well as the final session for the patient group, subjects took multiple measures of PTSD and general health and also underwent functional MRI (fMRI) scans while they did a combat picture viewing task (If you don’t know what fMRI is, it’s basically taking rapid-fire pictures of your brain every few seconds so you can see what the brain’s activity looks like at different points in time).
The critical manipulation in this experiment is the therapeutic training. The training is a series of four 5-minute tasks, each repeated twice. Three of these tasks were “affective labeling” tasks. In each of these three tasks, patients saw what we call “negatively valenced” images – that is, images associated with negative emotions – and were asked to choose from a list of emotionally charged words the one that was best associated with each image; that is, they were asked to name the emotions or emotionally charged aspects of the image on screen. One set of these images included specifically combat-related images, while the other two involved non-combat trauma-related images or negative facial expressions. In the final task, patients completed a classic Go-No Go task, in which they were instructed to press a button on each trial at the onset of a signal except when the signal specifically indicated not to. This latter task is what’s called a “response inhibition” task, because it requires you to overcome the initial reaction to press the button when you first see any signal appear onscreen.
It might seem odd at first to have patients complete a training for combat-related PTSD where just 25% of the training actually includes combat image exposure; how is labeling a sad face with the word “sad” supposed to reduced traumatic responses? The answer to this question is also why the therapeutic training includes the oddball response inhibition task among a set of affective labeling tasks. It all has to do with that word inhibition. PTSD is fundamentally understood to be a disorder in which patients cannot properly downregulate their emotions to certain trigger stimuli; that is, they cannot “inhibit” their emotional responses. The logic of the authors’ training structure is that the way the brain allows us to inhibit the button press during a No-Go task involves a similar type of circuit to the one that allows us to downregulate our emotions, and that affective labeling is a tool that can be used to engage that circuit as they have shown before in healthy patients.
The circuit at the heart of the authors’ thesis involves two primary brain areas. First is the amygdala, often considered the “fear center” of the brain. This area shows increased activity during increased fear or perceived threat, and often shows increased baseline activity (that is, its activity level at rest) in PTSD relative to healthy subjects. The second area is the “right ventrolateral prefrontal cortex”, or R-VLPFC, a mouthful of a brain area that has been associated with response inhibition. If you read last week’s Review Corner, you know the prefrontal cortex is what’s in front of your fingers if you put them up to your temples like elephant ears (no, I did not plan that). To understand where RVLPFC is, you’ll need to know that the bottom of the prefrontal cortex is roughly at your eyebrows. Ventro- (or “ventral”) in the context of the brain means toward your feet, and lateral means toward the skull, so the RVLPFC is that chunk along the outside from the lower part of your finger forward to about the middle of your eye on the right side. This latter region is key to the authors’ training regimen.
The RVLPFC is thought to be involved in inhibiting activity in other brain areas to downregulate responses that are contrary to your current goals. In response inhibition, it downregulates motor areas to keep you from pressing the button; for emotional inhibition, it downregulates the amygdala to keep you from acting on a threat response. However, in PTSD patients, we see reduced activity in this area, both during response inhibition and also when engaging with trauma responses. Together, we see increased amygdala response and decreased RVLPFC response in these patients, which suggests that at least part of the PTSD triggering is due to failed emotional response inhibition to certain emotional triggers.
Thus, the authors devised a training regimen to try engage the different inhibition circuits through response inhibition and affective labeling tasks, essentially “retraining” the RVLPFC. After collecting the data, they assessed two critical outcomes: the clinical scores, and the brain activity changes. Looking at patients’ overall PTSD level pre- and post-treatment, they found reduced scores for PTSD as well as depression symptoms, and of 12 participants who successfully completed all aspects of the study (including a later followup interview), 10 showed reduced symptoms and 5 were fully in remission. In the brain data, they found that PTSD scores were significantly correlated with amygdala activity during trauma image viewing, such that a bigger reduction (i.e., the scores went down more) in PTSD scores related to a bigger reduction (i.e., the activity went down more) in amygdala response from pre-treatment to post-treatment.
This study overall presents preliminary evidence that the authors’ affective labeling training, and by extension the amygdala-RVLPFC circuit as a training target, are promising avenues for combat-related and other types of PTSD treatment development. The key word here, of course, is preliminary – this is an early clinical effort with several critical limitations, most of which the authors themselves acknowledge. Of course, the sample size is small (check the footnotes for my thoughts there), and there is significant dropout in any multi-session study like this that limit the generalizability of the results. The authors also point out that their results don’t directly compare their treatment regimen with other established treatments, and that some aspects of those treatments may be overlapping. However, there are two critical issues I have with this study that, while I don’t think they preclude publication, I wished the authors had addressed directly.
First, as I pointed out earlier, this was a passive control design. My issue isn’t with the use of a passive control cohort per se, but the lack of post-treatment period followup with the healthy control subjects. There is no reason to believe these subjects will have the same results at two different periods in time, and indeed there may be reason to believe that they may show worsening symptoms as a cohort. Almost certainly, this decision was borne out of resource and bandwidth limitations, and the results were not dependent on differences between groups, so they do stand on their own. Accounting for changes in time with your control matters though, and will need to be assessed at some point with this treatment method for it to become viable.
The second issue is that, for all the background they dedicate to setting up the RVLPFC as a new avenue for treatment engagement, they don’t actually look at the brain data? This was the single biggest question mark in my mind, and it’s what pushed this from a “minor revisions” to a “major revisions” paper for me. It’s not that they didn’t find RVLPFC activity and so had no effects to report; they never mention it in the results at all. If the setup is that RVLPFC has reduced activity and the amygdala has increased activity for PTSD patients, why wouldn’t you look at both? Were I peer reviewing this, I would request that the authors return to their data and assess VLPFC activity changes (I’d actually expect them to look bilaterally, but that’s a topic for a different week) in the same way they report their amygdala results. I would want them to do this and report it even if the end results were non-significant, because that’s a critical component of their theoretical backing and needs to be discussed in context with the rest of the results.
All that said, I do like these results and I think it’s a cool reflection of how giving names to things can shape our understanding of and interactions with them. In this case, giving a name to a feeling, giving it an abstract representation, appears to activate neural circuits that protect us from the most volatile emotional responses to that feeling, and gives us room to interpret and assess that feeling and have control over how we react to it. The authors themselves note that this has long underscored folk understanding of psychology: the idea that talking about our feelings helps us overcome them. Training PTSD patients, and perhaps patients with related mental health issues, to engage in this regulatory circuit may give them an easy and accessible way to treat their condition, which would be a wonderful development in a world where mental health treatment can be both hard to come by and mistrusted by some of those who need it most.
Thanks for reading Brain Blast! Don’t forget to subscribe and then join us in the comments for this week’s reader discussion: Do you prefer the twice-a-week format or a single weekly longform post?
A very productive method of finding paywalled articles, for those interested, is to go to Google Scholar and paste in the title of the article. Often, if a PDF is available from a source, Google will link to it in the search results on the right. If that’s not there, you can also try clicking “All X Versions” if you see it in the bottom row of links of your search result; sometimes, authors will post their articles to a lab website or similar, which may show up in the expanded results. Failing that, you can also legitimately just email the corresponding author and ask for it (generally they are listed on the official journal page, even if there is a paywall). You should definitely not just go to sci-hub dot se or whatever proxy domains might be floating around and paste in your title or DOI reference (it won’t work for articles published after early 2021 anyway thanks to legal battles so you should totally just give up).
This is likely not a meaningful conflict of interest; the first author is the only one affiliated as either a current or former contractor, and no financial connection is noted in the disclosures.
There’s a common misconception by the public that small sample size automatically equals bad study; this is not true. While certainly in a perfect world with infinite time and resources (and subjects/patients that fit the study criteria) every study would be perfectly powered and spare no expense to run every test and control condition, that’s not our reality. Small studies are critical for testing early ideas to rule things out or hone hypotheses without breaking the bank or blocking progress overall. Additionally, in clinical contexts there’s a real need to “take what you can get” – patients may take medications or have concurrent conditions or be physically incapable of performing some tasks, but the research needs someone to study, so you work around it and account for your limitations in your report. All that to say while I’ll consider the sample size in my assessment of a study’s scientific merit, it is never and should not be a blanket reason to dismiss the study’s findings.