Uploaded November 2023 | Updated September 2026, 1 day ago
Abstract: Cross-lingual semantic parsing transfers parsing capability from a high-resource language (e.g., English) to low-resource languages with scarce training data. Previous work has primarily considered silver-standard data augmentation or zero-shot methods, however, exploiting few-shot gold data is comparatively unexplored. We propose a new approach to cross-lingual semantic parsing by explicitly minimizing cross-lingual divergence between probabilistic latent variables using Optimal Transport. We demonstrate how this direct guidance improves parsing from natural languages using fewer examples and less training. We evaluate our method on two datasets, MTOP and MultiATIS++SQL, establishing state-of-the-art results under a few-shot cross-lingual regime. Ablation studies further reveal that our method improves performance even without parallel input translations. In addition, we show that our model better captures cross-lingual structure in the latent space to improve semantic representation similarity. Given the improvement in semantic structure, we also consider new areas for improvement in cross-lingual structured alignment and how to approach future challenges in this domain. arxiv.org/abs/2307.04096
Bio: I am a final year PhD student at Edinburgh advised by Mirella Lapata on data-efficient cross-lingual semantic parsing and structure prediction. My research focuses on applying mathematical modeling to overcome data and resource constraints in adapting neural models to languages beyond English. My recent work focuses on optimization strategies for this goal including flatness-seeking optimization robustness and compute-efficient meta-learning. I was also recently awarded Outstanding Paper at ACL 2023 for collaborative work on MT evaluation. I'm excited by methods to improve data, resource and parameter efficiency for both cross-language adaptation and wider modeling improvement. I have previously interned at the Allen Institute for AI and Apple Siri. Prior to my PhD, I completed Masters degrees at University of Edinburgh, University of Cambridge and University College London.
tomsherborne.github.io
Abstract: Cross-lingual semantic parsing transfers parsing capability from a high-resource language (e.g., English) to low-resource languages with scarce training data. Previous work has primarily considered silver-standard data augmentation or zero-shot methods, however, exploiting few-shot gold data is comparatively unexplored. We propose a new approach to cross-lingual semantic parsing by explicitly minimizing cross-lingual divergence between probabilistic latent variables using Optimal Transport. We demonstrate how this direct guidance improves parsing from natural languages using fewer examples and less training. We evaluate our method on two datasets, MTOP and MultiATIS++SQL, establishing state-of-the-art results under a few-shot cross-lingual regime. Ablation studies further reveal that our method improves performance even without parallel input translations. In addition, we show that our model better captures cross-lingual structure in the latent space to improve semantic representation similarity. Given the improvement in semantic structure, we also consider new areas for improvement in cross-lingual structured alignment and how to approach future challenges in this domain. arxiv.org/abs/2307.04096
Bio: I am a final year PhD student at Edinburgh advised by Mirella Lapata on data-efficient cross-lingual semantic parsing and structure prediction. My research focuses on applying mathematical modeling to overcome data and resource constraints in adapting neural models to languages beyond English. My recent work focuses on optimization strategies for this goal including flatness-seeking optimization robustness and compute-efficient meta-learning. I was also recently awarded Outstanding Paper at ACL 2023 for collaborative work on MT evaluation. I'm excited by methods to improve data, resource and parameter efficiency for both cross-language adaptation and wider modeling improvement. I have previously interned at the Allen Institute for AI and Apple Siri. Prior to my PhD, I completed Masters degrees at University of Edinburgh, University of Cambridge and University College London.
tomsherborne.github.io
