Multichannel Generative Language Model: Learning All Possible Factorizations Within and Across Channels

TL;DR

The paper introduces MGLM, a flexible multichannel generative model that learns all possible factorizations across multiple languages, enabling diverse inference tasks and outperforming traditional models on multilingual data.

Contribution

MGLM is the first model to jointly learn and marginalize over all factorizations within and across multiple language channels.

Findings

MGLM successfully performs unconditional, conditional, and partially conditional generation.

It outperforms traditional bilingual discriminative models in quality-diversity trade-offs.

Qualitative samples demonstrate the model's ability to generate diverse multilingual outputs.

Abstract

A channel corresponds to a viewpoint or transformation of an underlying meaning. A pair of parallel sentences in English and French express the same underlying meaning, but through two separate channels corresponding to their languages. In this work, we present the Multichannel Generative Language Model (MGLM). MGLM is a generative joint distribution model over channels. MGLM marginalizes over all possible factorizations within and across all channels. MGLM endows flexible inference, including unconditional generation, conditional generation (where 1 channel is observed and other channels are generated), and partially observed generation (where incomplete observations are spread across all the channels). We experiment with the Multi30K dataset containing English, French, Czech, and German. We demonstrate experiments with unconditional, conditional, and partially conditional generation. We provide qualitative samples sampled unconditionally from the generative joint distribution. We also quantitatively analyze the quality-diversity trade-offs and find MGLM outperforms traditional bilingual discriminative models.