Language coverage and generalization in RNN-based continuous sentence embeddings for interacting agents

TL;DR

This paper evaluates the language coverage and generalization of RNN-based sentence embeddings using a CFG-generated dataset, revealing their limitations, and introduces a new arithmetic coding-based embedding method, AriEL, as a baseline.

Contribution

It provides a systematic evaluation of RNN embeddings' ability to cover language space and proposes AriEL, a non-data-driven embedding method based on arithmetic coding.

Findings

RNN embeddings underfit training data and cover limited language space.

They fail to learn the underlying CFG and generalize poorly to unbiased sentences.

AriEL offers an effective baseline for sentence encoding.

Abstract

Continuous sentence embeddings using recurrent neural networks (RNNs), where variable-length sentences are encoded into fixed-dimensional vectors, are often the main building blocks of architectures applied to language tasks such as dialogue generation. While it is known that those embeddings are able to learn some structures of language (e.g. grammar) in a purely data-driven manner, there is very little work on the objective evaluation of their ability to cover the whole language space and to generalize to sentences outside the language bias of the training data. Using a manually designed context-free grammar (CFG) to generate a large-scale dataset of sentences related to the content of realistic 3D indoor scenes, we evaluate the language coverage and generalization abilities of the most common continuous sentence embeddings based on RNNs. We also propose a new embedding method based on arithmetic coding, AriEL, that is not data-driven and that efficiently encodes in continuous space any sentence from the CFG. We find that RNN-based embeddings underfit the training data and cover only a small subset of the language defined by the CFG. They also fail to learn the underlying CFG and generalize to unbiased sentences from that same CFG. We found that AriEL provides an insightful baseline.