We often assume that language learning is primarily about words, grammar, & meaning—but what if our bodies are just as crucial to understanding concepts as our minds?
The Study:
A comprehensive study by researchers at Ohio State University & Hong Kong Polytechnic University put this to the test. They wanted to know if large language models like ChatGPT can truly understand human concepts without sensory experience. The team tested four major AI systems (GPT-3.5, GPT-4, PaLM, & Gemini) against human understanding across 4,442 English words.
The methodology was rigorous: researchers used two established databases the Glasgow Norms & the Lancaster Sensorimotor Norms, which collected data from thousands of human participants. Words were evaluated across various dimensions, from abstract qualities like emotional tone & concreteness to specific sensory experiences (visual, auditory, touch, smell & taste) & motor concepts (actions involving the hands, feet, or mouth).
Each AI model completed four rounds of testing, answering questions identical to human instructions, such as: “How arousing do you think ‘flower’ is on a 9-point scale?” or “How much do you experience ‘running’ through actions from your legs?”
The Findings:
The results revealed a striking pattern. For abstract qualities like emotional tone, the AI models showed strong correlations with human ratings. However, their performance dropped significantly for sensory dimensions & plummeted for motor concepts. The gap was most pronounced for action-related words—AI models correlated only weakly with human ratings for movements involving hands, feet, or mouths.
Statistical analysis confirmed this wasn’t a coincidence: all models showed significantly weaker alignment with human understanding as concepts moved from abstract to embodied domains. Intriguingly, models trained on both text & images (GPT-4 & Gemini) performed better on visually-related dimensions. This multimodal advantage suggests that additional input from images can help AI bridge some of the sensory gap.
What This Means:
This research strikes at the heart of longstanding debates about how we learn & understand language. The findings suggest that our physical bodies aren’t just vehicles for communication—they’re fundamental to how we construct meaning itself.
Consider how we naturally describe abstract concepts through physical metaphors: we “grasp” difficult ideas, feel “heavy” with sadness, or have “sharp” insights. These aren’t just linguistic decorations; they reflect how our embodied experiences shape our understanding. When we teach students that someone has a “warm” personality or a “cold” demeanour, we’re drawing on sensory experiences of temperature that AI systems, lacking bodies, cannot truly comprehend.
The implications extend beyond vocabulary to how we conceptualise learning itself. If sophisticated AI systems trained on vast amounts of text still struggle to understand embodied concepts, this suggests that physical experience isn’t just helpful for language learning—it may be essential for deep conceptual understanding.
The study’s findings appear to align with embodied cognition theory, which argues that our thinking is fundamentally shaped by our physical interactions with the world. For language learners, this means that engaging multiple senses—seeing, hearing, touching, & moving—may not just make learning more memorable but more meaningful. This research also illuminates why gesture is so powerful in language teaching. When students physically enact concepts—miming actions, using their hands to show size or direction, or moving through space to understand prepositions—they’re not just making abstract concepts concrete. They’re engaging the same embodied neural networks that ground their understanding in their native language.
If cutting-edge AI systems struggle to grasp embodied aspects of meaning despite processing vast amounts of linguistic data, this reveals something profound about the irreplaceable role of physical experience in human language acquisition. Our students don’t just think with their brains—they think with their entire bodies.
Teacher Takeaways:
- Incorporate multisensory vocabulary activities. Obviously, this works for action verbs and concrete nouns, but how about using it for metaphorical language that draws on physical experience. For example, when teaching “feeling blue,” you could have students look at or recall (or draw) something blue, or when discussing a “heavy heart,” you could ask them to hold a heavy object. Anyone fancy a bit of TPR?
- Recognise that your students’ physical experiences are valuable resources for language learning, not just their cognitive abilities.
What role does physicality play in your own language teaching or learning?
Leave a Reply