# Improving Semantic Composition with Offset Inference

## TL;DR

This paper introduces a novel distributional inference method leveraging type structures in Anchored Packed Trees to address sparsity in semantic models, enhancing their ability to infer plausible co-occurrences.

## Contribution

It presents a new inference technique that exploits type information in APTs to improve semantic composition and reduce data sparsity issues.

## Key findings

- Improved inference of plausible co-occurrences in semantic models.
- Enhanced performance of APTs in semantic composition tasks.
- Reduction of sparsity problems in distributional semantic models.

## Abstract

Count-based distributional semantic models suffer from sparsity due to unobserved but plausible co-occurrences in any text collection. This problem is amplified for models like Anchored Packed Trees (APTs), that take the grammatical type of a co-occurrence into account. We therefore introduce a novel form of distributional inference that exploits the rich type structure in APTs and infers missing data by the same mechanism that is used for semantic composition.