Tensor Product Representations of Regular Transductions


This paper provides a vector space characterization of regular transductions. We use finite model theory to characterize objects like strings and trees as relational structures and origin graphs to characterize input-output relations generated by transducer. We show detailed processes of using multilinear maps as function application for evaluation to compile regular transductions characterized by MSO definable origin graphs into a tensor embedding.

Proceedings of SCiL 2024
Jon Rawski
Jon Rawski
Assistant Professor of Linguistics