Transferring Neural Potentials For High Order Dependency Parsing

Dependency parsing is the basis of many complex pipelines for problems in natural language processing such as machine summarization, machine translation, event extraction, semantic parsing,semantic role labeling(SRL), emotion analysis, dialogue systems and information processing. Thus, any error in dependency parsing could propagate to downstream task and therefore any advance in this field could lead to major improvement in NLP tasks. There are two main approaches to dependency parsing. The first approach is transition based which has incremental local inference and involves using datastructures such as buffer and stack (Nivre 2008),(Buys & Blunsom 2015). This approach has the limitation of resolving relatively short sentences and is a trade-off between speed and accuracy.