A non-projective greedy dependency parser with bidirectional LSTMs

Abstract

The LyS-FASTPARSE team presents BIST-COVINGTON, a neural implementation of the Covington (2001) algorithm for non-projective dependency parsing. The bidirectional LSTM approach by Kiperwasser and Goldberg (2016) is used to train a greedy parser with a dynamic oracle to mitigate error propagation. The model participated in the CoNLL 2017 UD Shared Task. In spite of not using any ensemble methods and using the baseline segmentation and PoS tagging, the parser obtained good results on both macro-average LAS and UAS in the big treebanks category (55 languages), ranking 7th out of 33 teams. In the all treebanks category (LAS and UAS) we ranked 16th and 12th. The gap between the all and big categories is mainly due to the poor performance on four parallel PUD treebanks, suggesting that some 'suffixed' treebanks (e.g. Spanish-AnCora) perform poorly on cross-treebank settings, which does not occur with the corresponding 'unsuffixed' treebank (e.g. Spanish). By changing that, we obtain the 11th best LAS among all runs (official and unofficial). The code is made available at https://github.com/CoNLL-UD-2017/LyS-FASTPARSE

Untitled

“…There may indeed be several sources of inconsistencies in the gold annotations : in addition to the divergences in the theoretical linguistic principles that governed the design of the original annotation guidelines, inconsistencies may also result from automatic (pre-)processing, human post-editing, or human annotation. Actually, several studies have recently pointed out that treebanks for the same language are not consistently annotated (Vilares and Gómez-Rodríguez, 2017;Aufrant et al, 2017). In a closely related context, Wisniewski et al (2014) have also shown that, in spite of common annotation guidelines, one of the main bottleneck in cross-lingual transfer between UD corpora is the difference in the annotation conventions across treebanks and languages.…”

Comparing neural‐ and N‐gram‐based language models for word segmentation

“…The choice of using a recurrent neural network is based on its focus on dealing with sequential data, such as text, as well as its wide use in several NLP tasks, such as machine translation (Johnson et al, ), dependency parsing (Vilares & Gómez‐Rodríguez, ), question answering (Iyyer, Boyd‐Graber, Claudino, Socher, & Daumé III, ), or language modeling (Sundermeyer, Schlüter, & Ney, ). Recurrent neural networks differ from traditional feedfoward networks in that they allow feedback loops in their architectures, thus being able to use the output information corresponding to the input t when processing input t + 1 .…”