Dr. Vishnu Sreekumar and Anuska Maity, a just-graduated MS-by-research (CSE) student published their research work on Functional role and semantic structure shape word memorability: Cue–target asymmetries in cued recall in speciality journal – Cognition. Anuska researched the phenomenon of “memorability”, i.e., what makes some experiences more memorable than others to most people regardless of their individual differences? The literature often claims that memorability is a property that is intrinsic to the stimulus. This claim results from the fact that memorability scores are obtained by averaging memory performance for a given word or image (depending on what the stimulus is) over a large number of study participants, washing out individual differences, leaving only stimulus-dependent factors.

However, Anuska showed that the same word can take on different memorability depending on its role in the memory task. Specifically, in cued recall, a standard memory task used in the laboratory, a word can be used to cue memory but also be the target of memory search. Anuska showed that the same word does not retain its memorability very well across these roles. She also developed statistical and ML models based on psycholinguistic and semantic network properties to explain these discrepancies. A practical takeaway from this demonstration is quite simple: there are efforts today to develop ML models to predict memorability and to use such predictions in domains like advertising. However, those models are built using training data that come from recognition memory tasks (i.e., given a brand name, do you remember seeing that brand name before? Yes or no?). Anuska’s results imply that we need to do better: develop models to predict cue efficacy if the goal is to enhance the likelihood that a brand icon will cue the brand name. Similarly, train models to predict the ease of retrieving a concept if the goal is to enhance the retrievability of a brand name when cued with an effective brand icon.

In other ongoing work in the lab, they are expanding this line of work further by showing that memorability for the same stimuli varies significantly when tested using different memory test formats and even with the same test format but with the words translated to Hindi (i.e., changing the representational substrate). They are also moving beyond the level of words to sentences. They have previously also published work on video memorability, in collaboration with Dr. Makarand Tapaswi’s lab. Taken together, this body of work illustrates that memorability is not a unitary property that is intrinsic to stimuli but rather something that is determined by the specific cognitive demands of the memory task and the representational formats used by the resulting cognitive processes.

Full paper: https://www.sciencedirect.com/science/article/abs/pii/S0010027726001599?via%3Dihub=

July 2026