One Semantic Parser to Parse Them All: Sequence to Sequence Multi-Task Learning on Semantic Parsing Datasets

Semantic parsers map natural language utterances to meaning representations.\nThe lack of a single standard for meaning representations led to the creation\nof a plethora of semantic parsing datasets. To unify different datasets and\ntrain a single model for them, we investigate the use of Multi-Task Learning\n(MTL) architectures. We experiment with five datasets (Geoquery, NLMaps, TOP,\nOvernight, AMR). We find that an MTL architecture that shares the entire\nnetwork across datasets yields competitive or better parsing accuracies than\nthe single-task baselines, while reducing the total number of parameters by\n68%. We further provide evidence that MTL has also better compositional\ngeneralization than single-task models. We also present a comparison of task\nsampling methods and propose a competitive alternative to widespread\nproportional sampling strategies.\n