Error Analysis for Vietnamese Dependency Parsing

Abstract

Dependency parsing is needed in different applications of natural language processing. In this paper, we present a thorough error analysis for dependency parsing for the Vietnamese language, using two state-of-the-art parsers: MSTParser and MaltParser. The error analysis results provide us insights in order to improve the performance of dependency parsing for the Vietnamese language.

LSTM Easy-first Dependency Parsing with Pre-trained Word Embeddings and Character-level Word Embeddings in Vietnamese

“…In our experiments, we use the VnDT [15] which has 10,200 dependency structures. For comparison purposes, we split the data in the same way as [18], last 1020 sentences for testing (POS and automatic POS tagging) and the rest for training. The POS tagging tool we used is VnTagger 1 , its accuracy is 94.4% on our test set.…”

“…We found that [18] did not incorporate pre-trained word embeddings in the vector representation of words. Although this feature can improve the accuracy of dependency parsing because of its rich context information [11] [12] and [17].…”

“…Nguyen and Nguyen [16] used supertags features and Vu-Manh et al [17] used word embeddings on the transitionbased parser. Nguyen et al [18] used BiLSTM encoder to generate word representation vectors and obtained a state-ofthe-art result on Vietnamese Dependency Treebank (VnDT) [15].…”

A Neural Network Model for Efficient Antonymy-Synonymy Classification by Exploiting Co-occurrence Contexts and Word-Structure Patterns

“…Therefore, Mandar Joshi and co-authors [22] proposed pair2vec model exploiting surface form of contexts instead of dependency paths. From the review of previous studies in subsection 1.1, there are still two open problems as follows: -Using neural network models to exploit cooccurrence contexts of word pairs without relying on syntactic parse trees: For low-resource languages like Vietnamese, syntactic parsers can only achieve an accuracy much lower than those for English [23]. Table 1 shows a number of word length statistics from a popular Vietnamese dictionary conducted by Nguyen and co-authors [24].…”

Vietnamese transition-based dependency parsing with supertag features

“…According to the error analysis results for Vietnamese dependency parsing in [12], dependency parsing tends to have lower accuracies on long dependencies. We can see that a lot of dependency relations (noun modifier, verb modifier, subject, direct object, root, adjective modifier, coordination, conjunction, indirect object) have lower accuracies than others, as shown in Table I (the figures in this table are from the paper [12]). Therefore, we focus on finding new features to solve above problems related to these dependency relations.…”