This is an implementation in Kotlin of the parser described in Non-Projective Dependency Parsing via Latent Heads Representation (LHR), based on the SimpleDNN neural network library.

LHRParser is a neural dependency parser that implements a novel approach based on a bidirectional recurrent autoencoder to perform globally optimized non-projective parsing via semi-supervised learning.

The image shows the architecture, composed by two BiLSTM encoders that produce the Context Vectors and the Latent Heads, a Similarity decoder that finds the relations between dependents and governors and the Labeler that assigns a dependency relation label and a part-of-speech tag to each token.

If you make use of this software for research purposes, we'll appreciate citing the following:

@ARTICLE{2018arXiv180202116G, 
    author = {{Grella}, M. and {Cangialosi}, S.}, 
    title = "{Non-Projective Dependency Parsing via Latent Heads Representation (LHR)}", 
    journal = {ArXiv e-prints}, 
    archivePrefix = "arXiv", 
    eprint = {1802.02116}, 
    primaryClass = "cs.CL", 
    keywords = {Computer Science - Computation and Language}, 
    year = 2018, 
    month = feb, 
    adsurl = {http:https://adsabs.harvard.edu/abs/2018arXiv180202116G}, 
    adsnote = {Provided by the SAO/NASA Astrophysics Data System}
}

For questions and usage issues, please contact [[email protected]].