Siamese Networks for Semantic Pattern Similarity

TL;DR

This paper introduces a Siamese Network approach to measure semantic pattern similarity between sentences, focusing on abstract patterns rather than specific meanings, with applications in database question answering.

Contribution

It presents a novel application of Siamese Networks for semantic pattern similarity, demonstrating high accuracy and confidence estimation capabilities.

Findings

Achieves high accuracy in semantic pattern similarity tasks.

Provides a confidence measure for predictions.

Effective in SQL pattern determination for unseen questions.

Abstract

Semantic Pattern Similarity is an interesting, though not often encountered NLP task where two sentences are compared not by their specific meaning, but by their more abstract semantic pattern (e.g., preposition or frame). We utilize Siamese Networks to model this task, and show its usefulness in determining SQL patterns for unseen questions in a database-backed question answering scenario. Our approach achieves high accuracy and contains a built-in proxy for confidence, which can be used to keep precision arbitrarily high.