Cross-Task Knowledge-Constrained Self Training

TL;DR

This paper introduces a framework for multi-task learning that leverages prior knowledge about task output spaces, with theoretical PAC analysis and practical improvements in shallow parsing and NER tasks.

Contribution

It proposes a novel multi-task learning framework utilizing output space relations, supported by PAC learning analysis and demonstrated through improved NLP task performance.

Findings

PAC learning conditions identified for multi-task frameworks

Consistent performance improvements in shallow parsing

Enhanced named-entity recognition results

Abstract

We present an algorithmic framework for learning multiple related tasks. Our framework exploits a form of prior knowledge that relates the output spaces of these tasks. We present PAC learning results that analyze the conditions under which such learning is possible. We present results on learning a shallow parser and named-entity recognition system that exploits our framework, showing consistent improvements over baseline methods.