VPCTagger: Detecting Verb-Particle Constructions With Syntax-Based Methods

Abstract

Verb-particle combinations (VPCs) consist of a verbal and a preposition/particle component, which often have some additional meaning compared to the meaning of their parts. If a data-driven morphological parser or a syntactic parser is trained on a dataset annotated with extra information for VPCs, they will be able to identify VPCs in raw texts. In this paper, we examine how syntactic parsers perform on this task and we introduce VPCTagger, a machine learning-based tool that is able to identify English VPCs in context. Our method consists of two steps: it first selects VPC candidates on the basis of syntactic information and then selects genuine VPCs among them by exploiting new features like semantic and contextual ones. Based on our results, we see that VPCTagger outperforms state-of-the-art methods in the VPC detection task.

Verb-Particle Constructions in Questions

“…we did not carry out any filtering steps). For the sake of comparability with previous results obtained for this corpus (Tu and Roth, 2012;Nagy T. and Vincze, 2014), here we applied an SVM model (Cortes and Vapnik, 1995) with 5 fold cross validation and obtained an accuracy score of 80.05%. On the same data, Tu and Roth (2012) obtained an accuracy score of 78.6%, which was outperformed by Nagy T. and Vincze (2014) with a score of 81.92%.…”

Proceedings of the 13th Workshop on Multiword Expressions (MWE 2017)

“…Previous approaches for VMWE identification include the two-pass method of candidate extraction followed by binary classification (Fazly et al, 2009;Nagy T. and Vincze, 2014). VMWE identification has also been performed using sequence labeling approaches, with IOBscheme.…”

“…First of all, syntactic structure is known to help MWE identification (Fazly et al, 2009;Seretan, 2011;Nagy T. and Vincze, 2014). We therefore inform the system with the provided syntactic dependencies when available: for each token B n that both appears in the buffer and is a syntactic dependent of S 0 with label l, we capture the existence of the dependency using the features RightDep(S 0 , B n ) = T rue and RightDepLabel(S 0 , B n ) = l. We also use the opposite features IsGovernedBy(S 0 , B n ) = T rue and IsGovernedByLabel(S 0 , G) = l when S 0 's syntactic governor G appears in the buffer.…”

“…Tu and Roth (2012) used syntactic and lexical features to classify VPCs candidates on a crowdsourced corpus. Nagy T. and Vincze (2014) implemented VPCTagger, a machine learning-based tool that selects VPC candidates on the basis of syntactic information and then classifies them as VPCs or not, based on lexical, syntactic and semantic features. Smith (2014) extracted VPCs from an English-Spanish parallel subtitles corpus.…”

“…We implemented some of the features used by Nagy T. and Vincze (2014) and based on their results, we trained a J48 model (Quinlan, 1993) and an SVM model (Cortes and Vapnik, 1995) on the data (using Weka's ) default settings) applying ten fold cross validation. As an evaluation metric, we used accuracy score.…”