Annotating Chinese Light Verb Constructions according to PARSEME guidelines

Annotating Chinese Light Verb Constructions according to PARSEME guidelines Menghan Jiang, Natalia Klyueva, Hongzhi Xu and Chu-Ren Huang The Hong Kong Polytechnic University University of Pennsylvania menghan.jiang, natalia.klyueva, churen.huang@polyu.edu.hk xh@cis.upenn.edu Abstract In this paper we present a preliminary study on application of PARSEME guidelines of annotating mutiword expressions to Chinese language. We focus on one specific category - light verb constructions (LVCs). We make use of an existing resource containing Chinese light verbs and examine whether this resource fulfill the requirements of the guidelines. We make a preliminary annotation of a Chinese UD treebank in two steps: first automatically identifying potential light verbs and then manually assigning the corresponding nouns or correcting false positives. Keywords: verbal multiword expressions, light verb constructions, Chinese, corpus linguistics 1. Introduction Multiword Expressions (MWEs) present a challenge in a bunch of areas of Natural Language Processing like Ma- chine Translation or Information Extraction. They are id- iosyncratic in their nature - the meaning of the whole can not be derived from the meaning of its parts. The exact def- inition of MWEs varies across different linguistic theories and can not be necessarily universal for all languages. PARSEME (Savary et al., 2017) is a European project which aims at processing multiword expressions from dif- ferent perspectives and for various languages. Universal guidelines were created to annotate verbal multiword ex- pressions distinguishing several subtypes unique for a lan- guage: idioms, LVCs, verb-particle constructions, inher- ently reflexive verbs and others (miscellaneous category). Corpora in 18 languages have been annotated which result in a multilingual resource (Savary et al., 2017). One of the long-term plans of the project is to extend the set of (mostly European) languages to Asian languages, includ- ing Chinese Mandarin. Chinese linguistic tradition is dif- ferent from the European which presents a challenge when trying to apply the guidelines created under the European project to Chinese. In this paper we make a pilot study of one particular type of verbal MWEs - LVCs and see how the existing resources for Chinese can be adjusted to the PARSEME annotation schema. We consider two possible corpora for annotation: the Sinica Balanced Corpus of Modern Chinese(Huang et al., 2000) and Universal Dependencies (Nivre et al., 2017). Fi- nally, we have chosen Chinese Universal Dependencies treebank1 (wiki data) because it features syntactic anno- tation in dependency-based format required by the shared task.Unlike other languages such as Czech (Bejˇcek et al., 2017), there is no resources we can use to generate MWEs automatically for Chinese. But we don’t do the annotation from scratch as well. We use a list of Chinese light verbs to pre-process the data, and then manual work is performed to 1http://universaldependencies.org/zh/ overview/introduction.html assign the corresponding nouns of the MWEs and exclude the false ones. This paper is structured as follows. In Section 2 we de- scribe which types of verbal MWEs can occur in Chinese, in Section 3 we give more extensive information on Chi- nese LVCs, and in Section 4 we explore how the annotation …

Content truncated to 500 words