Functorial Language Models

This work addresses a challenge in computational linguistics for researchers by providing a principled method for training DisCoCat models, though it appears incremental as it builds on existing categorical frameworks.

The authors tackled the problem of training categorical compositional distributional (DisCoCat) models on raw text data by introducing functorial language models, which compute probability distributions over word sequences using a monoidal functor from grammar to meaning, and they implemented a proof-of-concept in DisCoPy.