AR-Engine - a framework for unrestricted co-reference resolution


Dan Cristea ("Al.I.Cuza" University of Iasi, Romania)

Oana-Diana Postolache ("Al.I.Cuza" University of Iasi, Romania)

Gabriela-Eugenia Dima ("Al.I.Cuza" University of Iasi, Romania)

Catalina Barbu (University of Wolverhampton, UK)


WO22: Coreference


The paper presents a framework that allows the design, realisation and validation of different anaphora resolution models on real texts. The type of processing implemented by the engine is an incremental one, simulating the reading of texts by humans. Advanced behaviour like postponed resolution and accumulation of values for features of the discourse entities during reading is implemented. Four models are defined, plugged in the framework and tested on a small corpus. The approach is open to any type of anaphora resolution. However, the models reported deal only with co-reference anaphora, independent of the type of the anaphor. It is shown that the setting on of more and more features, generally results in an improvement of the analysis.


Anaphora resolution, Framework, Resolution model, Evaluation

Full Paper