1 paper
Zhijin Guo, Chenhao Xue, Zhaozhen Xu +4
For language models to generalize correctly to novel expressions, it is critical that they exploit access compositional meanings when this is justified. Even if we don't know what…