We often forget how tricky metaphor can be- until we try to teach it… or build a machine that understands it.
Metaphors aren’t just colourful language- they reflect how we think. We say things like time is running out or he exploded with anger, without thinking twice. But for computers, recognising what’s literal & what’s metaphorical is a big challenge.
That’s what makes this new study by Dongmei Zhu (2025) so interesting: Zhu designed a system that combines two types of AI tools -CNNs (Convolutional Neural Networks) & SVMs (Support Vector Machines)- to help computers detect metaphors more accurately. The model was tested on both English & Chinese datasets, including metaphor-rich verb phrases like “envy ate at him.”
Here’s what the system does:
- CNNs help the model spot patterns in sentences that might suggest a metaphor- for example, strange verb-object combinations.
- SVMs then decide whether those patterns really are metaphors.
- The model also looks at grammar (part-of-speech tags) & meaning information from WordNet (a large lexical dataset of English) to improve understanding.
Despite using small datasets, the results were strong- 85% accuracy for English verb metaphors, which outperformed even well-known models like RoBERTa & DeBERTa.
For me, what’s interesting about Zhu’s work is that it also draws on cognitive linguistics, the idea that metaphors shape how we think- not just how we speak. It connects language processing with how the brain links ideas across different areas (such as emotions & time).
What’s exciting is how this research shows AI can be trained to “read between the lines”- a skill that’s crucial in language, literature & communication.
Teacher Takeaways?
- Teach context, not just words: Like humans, AI performs better when it looks at the grammar & context of phrases. It’s a reminder to teach language in meaningful chunks, not isolated words.
- Some metaphors are harder than others: The model found verb metaphors easier to detect than adjective-noun ones. This mirrors what learners often experience too.
- Culture matters: The system worked across English & Chinese, showing how metaphors can vary between languages. It’s a good reason to explore how learners’ first languages influence metaphor use.
Even if this research doesn’t link directly to your next lesson plan, it’s a fascinating glimpse into how language, thought & AI are starting to overlap. It could also help build future tools for learning & analysing figurative language.
Do you teach metaphor explicitly—or let students discover it naturally through reading & listening?
Leave a Reply