Advancing NLP with Cognitive Language Processing Signals

  • 2019-04-04 17:38:16
  • Nora Hollenstein, Maria Barrett, Marius Troendle, Francesco Bigiolli, Nicolas Langer, Ce Zhang
  • 8

Abstract

When we read, our brain processes language and generates cognitive processingdata such as gaze patterns and brain activity. These signals can be recordedwhile reading. Cognitive language processing data such as eye-tracking featureshave shown improvements on single NLP tasks. We analyze whether using suchhuman features can show consistent improvement across tasks and data sources.We present an extensive investigation of the benefits and limitations of usingcognitive processing data for NLP. Specifically, we use gaze and EEG featuresto augment models of named entity recognition, relation classification, andsentiment analysis. These methods significantly outperform the baselines andshow the potential and current limitations of employing human languageprocessing data for NLP.

 

Introduction (beta)

None

 

Conclusion (beta)

None