Corpus ID: 212658007

A Benchmark for Systematic Generalization in Grounded Language Understanding

@article{Ruis2020ABF,
  title={A Benchmark for Systematic Generalization in Grounded Language Understanding},
  author={Laura Ruis and Jacob Andreas and Marco Baroni and Diane Bouchacourt and B. Lake},
  journal={ArXiv},
  year={2020},
  volume={abs/2003.05161}
}
Human language users easily interpret expressions that describe unfamiliar situations composed from familiar parts ("greet the pink brontosaurus by the ferris wheel"). Modern neural networks, by contrast, struggle to interpret compositions unseen in training. In this paper, we introduce a new benchmark, gSCAN, for evaluating compositional generalization in models of situated language understanding. We take inspiration from standard models of meaning composition in formal linguistics. Going… Expand
21 Citations
Think before you act: A simple baseline for compositional generalization
  • 1
  • PDF
Improving Compositional Generalization in Semantic Parsing
  • 5
  • PDF
Systematic Generalization on gSCAN with Language Conditioned Embedding
  • Highly Influenced
  • PDF
Compositional Generalization via Neural-Symbolic Stack Machines
  • 5
  • PDF
The EOS Decision and Length Extrapolation
  • 3
  • PDF
Word meaning in minds and machines
  • 3
  • PDF
...
1
2
3
...

References

SHOWING 1-10 OF 42 REFERENCES
Rearranging the Familiar: Testing Compositional Generalization in Recurrent Networks
  • 61
  • PDF
The compositionality of neural networks: integrating symbolism and connectionism
  • 17
  • PDF
Permutation Equivariant Models for Compositional Generalization in Language
  • 24
Generalization without Systematicity: On the Compositional Skills of Sequence-to-Sequence Recurrent Networks
  • 218
  • PDF
Emergent Systematic Generalization in a Situated Agent
  • 24
BabyAI: A Platform to Study the Sample Efficiency of Grounded Language Learning
  • 45
  • PDF
Learning Compositional Rules via Neural Program Synthesis
  • 14
  • PDF
...
1
2
3
4
5
...