“`html
“You still needed to demonstrate your worth.”
“Every cloud possesses a silver lining!”
Which of these statements are you most likely to recall in a few minutes? If you guessed the latter, you are likely correct.
Based on a recent investigation from MIT cognitive scientists, propositions that linger in your memory longer are those that deliver unique meanings, causing them to differentiate from phrases you’ve encountered before. They discovered that meaning, rather than any other attribute, is the most crucial aspect when it comes to memorability.
“One might have presumed that remembering sentences is largely about their visual characteristics, but our findings indicated otherwise. A significant contribution of this study is pinpointing that the meaning-related context renders sentences memorable,” states Greta Tuckute PhD ’25, currently a research fellow at Harvard University’s Kempner Institute.
The results bolster the theory that phrases with unique meanings — such as “Does olive oil assist with tanning?” — are stored in cognitive spaces that lack clutter from sentences with similar meanings. Phrases with overlapping meanings become densely grouped, making it more challenging to confidently recognize them later, according to the researchers.
“When you encode sentences that share a comparable meaning, there’s feature overlap within that space. Thus, a specific sentence you’ve encoded is not associated with a distinct set of features, but rather with a collection of features that may overlap with those of other sentences,” explains Evelina Fedorenko, an associate professor of brain and cognitive sciences (BCS) at MIT, a member of MIT’s McGovern Institute for Brain Research, and the senior author of the research.
Tuckute and Thomas Clark, a graduate student at MIT, serve as the primary authors of the paper, which is published in the Journal of Memory and Language. MIT graduate student Bryan Medina is also a co-author.
Unique sentences
What differentiates specific elements as being more memorable than others remains a longstanding inquiry in cognitive science and neuroscience. In a 2011 investigation, Aude Oliva, now a senior research scientist at MIT and director of the MIT-IBM Watson AI Lab, demonstrated that not all items hold the same significance: Certain types of images are considerably easier to recall than others, and people are remarkably consistent in recalling which images they find most memorable.
In that research, Oliva and her team discovered that, generally, images featuring people are the most unforgettable, followed closely by images of human-scale environments and close-ups of objects. The least memorable consist of natural landscapes.
As a follow-up to that study, Fedorenko, Oliva, and Ted Gibson, another faculty member in BCS, collaborated to ascertain whether words also vary in their memorability. In a study published earlier this year and jointly led by Tuckute and Kyle Mahowald, a former PhD student in BCS, the researchers discovered that the most memorable words are those that possess the most unique meanings.
Words are deemed more distinctive if they have a singular meaning and few or no synonyms — examples being “pineapple” or “avalanche,” which were found to be highly memorable. Conversely, words with multiple meanings, such as “light,” or those with numerous synonyms, like “happy,” proved more challenging for individuals to recognize accurately.
In the recent study, the researchers broadened their focus to evaluate the memorability of sentences. Just as with words, some phrases convey very unique meanings, while others express similar information in slightly varied manners.
To conduct the study, the researchers compiled a collection of 2,500 sentences sourced from publicly available databases compiling text from novels, news articles, movie dialogues, and additional sources. Every sentence they selected contained precisely six words.
The researchers then showcased a random selection of around 1,000 of these sentences to each participant in the study, including some repeats. Each of the 500 participants was instructed to press a button when they encountered a sentence they recognized from earlier.
The most unforgettable sentences — those where participants quickly and accurately indicated they had seen them before — comprised phrases such as “Homer Simpson is hungry, very hungry,” and “These mosquitoes are — well, guinea pigs.”
Those memorable sentences corresponded significantly with phrases deemed to possess unique meanings, as estimated through the vast vector space of a large language model (LLM) called Sentence BERT. This model can generate sentence-level representations of phrases, which are useful for tasks such as assessing meaning similarity between sentences. The model provided researchers with a distinctness score for each sentence based on its semantic resemblance to other sentences.
The researchers also evaluated the sentences using a model that predicts memorability based on the average memorability of the individual words within the sentence. This model performed reasonably well at estimating overall sentence memorability, but not as effectively as Sentence BERT. This indicates that the meaning of a sentence as a whole — beyond the contributions from individual words — dictates its memorability, the researchers assert.
Chaotic memories
While cognitive scientists have long suggested that the brain’s memory reserves are finite, the results of the new study endorse an alternative hypothesis that may help clarify how the brain can continue to form new memories without sacrificing old ones.
This alternative, termed the noisy representation hypothesis, posits that when the brain encodes a new memory—be it an image, a word, or a sentence—it is represented in a noisy manner. This suggests that the representation is not identical to the stimulus, and some information is lost. For instance, in the case of an image, you might not encode the precise angle at which an object is viewed, and for a sentence, you might not remember the exact structure employed.
Under this theory, a new sentence would be encoded in a similar cognitive space as sentences that convey similar meanings, whether they were encountered recently or at any other point within a lifetime of linguistic experiences. This intermingling of similar meanings raises the level of noise and can make it significantly more challenging to remember the precise sentence you’ve seen before.
“The representation is gradually accruing some noise. Consequently, when you view an image or a sentence for a second time, your accuracy in judging whether you’ve encountered it before will be impacted, and it will be lower than 100 percent in most instances,” Clark states.
However, if a sentence has a unique meaning that is encoded in a less densely populated space, it will be simpler to identify later.
“Your memory may still be noisy, but your capacity to make judgments based on the representations is less compromised by that noise because the representation is inherently distinctive,” Clark explains.
The researchers now intend to investigate whether additional features of sentences, such as more vivid and descriptive language, may also contribute to their memorability, and how the language system may interact with hippocampal memory structures during the encoding and retrieval of memories.
The research was partially funded by the National Institutes of Health, the McGovern Institute, the Department of Brain and Cognitive Sciences, the Simons Center for the Social Brain, and the MIT Quest Initiative for Intelligence.
“`