# Latent Variable Sentiment Grammar

## TL;DR

This paper introduces a sentiment grammar framework that explicitly models sentiment composition using latent variables and Gaussian mixtures, improving neural sentiment classifiers on the SST benchmark.

## Contribution

It proposes two formal models for deep sentiment representation that explicitly encode sentiment subtypes, enhancing neural sentiment analysis.

## Key findings

- Sentiment grammar outperforms vanilla neural encoders.
- Using ELMo embeddings yields state-of-the-art results.
- Models effectively capture sentiment subtype expressions.

## Abstract

Neural models have been investigated for sentiment classification over constituent trees. They learn phrase composition automatically by encoding tree structures but do not explicitly model sentiment composition, which requires to encode sentiment class labels. To this end, we investigate two formalisms with deep sentiment representations that capture sentiment subtype expressions by latent variables and Gaussian mixture vectors, respectively. Experiments on Stanford Sentiment Treebank (SST) show the effectiveness of sentiment grammar over vanilla neural encoders. Using ELMo embeddings, our method gives the best results on this benchmark.