Deep Neural Networks and Humans Benefit from Compositional Language Structure
Author Information
Author(s): Galke Lukas, Ram Yoav, Raviv Limor
Primary Institution: University of Southern Denmark
Hypothesis
Do deep neural network models exhibit the same learning and generalization advantage when trained on more structured linguistic input as human adults?
Conclusion
The study demonstrates that deep neural networks, like humans, show a learnability advantage when trained on languages with more structured linguistic input.
Supporting Evidence
- Neural networks exposed to more compositional languages show more systematic generalization.
- Greater agreement between different agents was observed when the input was more compositional.
- Neural networks trained on highly structured languages produced more transparent generalizations.
Takeaway
This study shows that both humans and deep neural networks learn languages better when those languages have clear and structured rules.
Methodology
The study compared the learning and generalization capabilities of deep neural networks and humans using ten input languages with varying degrees of compositional structure.
Participant Demographics
Adult human participants were involved in the study.
Statistical Information
P-Value
p<0.001
Statistical Significance
p<0.001
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website