Korean Coreference Resolution with Guided Mention Pair Model Using Deep Learning

Abstract

The general method of machine learning has encountered disadvantages in terms of the significant amount of time and effort required for feature extraction and engineering in natural language processing. However, in recent years, these disadvantages have been solved using deep learning. In this paper, we propose a mention pair (MP) model using deep learning, and a system that combines both rule‐based and deep learning‐based systems using a guided MP as a coreference resolution, which is an information extraction technique. Our experiment results confirm that the proposed deep‐learning based coreference resolution system achieves a better level of performance than rule‐ and statistics‐based systems applied separately.

Mention Detection Using Pointer Networks for Coreference Resolution

“…We measured the performance of rule-based coreference resolution based on a model showing the optimal results of the mention detection using CoNLL F1 [1]- [3] (see Table 14). Tables 4-12 show the tuning of the hyperparameters in the experiments.…”

“…First, [1], [2] extracted mentions using a left-toright, breadth-first search based on phrase structure analysis and certain rules. Then, [3] extracted nouns or noun phrases as a head based on a dependency tree, and mention detection defined a mention boundary using a modifier and head. Rule-based mention detection can extract the correct mentions to define the appropriate boundary for a relatively short noun phrase such as a normally single noun, compound noun, or head with a short-range modifier.…”

“…Rule-based mention detection is performed based on syntactic analysis and certain rules defined by humans [1]- [3]. First, [1], [2] extracted mentions using a left-toright, breadth-first search based on phrase structure analysis and certain rules.…”

“…We can show an increase in the recall of coreference resolution with more mentions because the mention is the unit of any entity (that is, the subject) in the coreference resolution. Standard mention detection was processed using rule-based [1]- [3] and statistics-based [7]- [13] approaches.…”

“…Coreference resolution is based on mentions, and the method for extracting mentions in a document is known as mention detection [1]- [3].…”