Paraphrasing of Italian Support Verb Constructions based on Lexical and Grammatical Resources

Abstract

Support verb constructions (SVC), are verb-noun complexes which play a role in many natural language processing (NLP) tasks, such as Machine Translation (MT). They can be paraphrased with a full verb, preserving its meaning, improving at the same time the MT raw output. In this paper, we discuss the creation of linguistic resources namely a set of dictionaries and rules that can identify and paraphrase Italian SVCs. We propose a paraphrasing computational method that is based on open-source tools and data such as NooJ linguistic environment and OpenLogos MT system. We focus on pre-processing the data that will be machine translated, but our methodology can also be applied in other fields in NLP. Our results show that linguistic knowledge constitutes a 95.5% precision rate in identifying SVC and an 88.8% precision rate in paraphrasing SVCs into full verbs.

Recognizing Verb-Based Croatian Idiomatic MWUs

“…the verb never changes gender, number, person or tense, are placed in the category 1 (nije šala, ni pas s maslom ne bi pojeo, povuci-potegni, pričam ti priču). They are listed in the dictionary as simple entries (Kocijan, Librenjak, to appear) 2 and belong to fixed expressions, as termed by Sag et al (2002).…”

“…Provided that we assure the detailed syntactic description and categorization of idioms, NooJ can be used to construct grammars to detect any known idioms in a Croatian text. Using somewhat different approaches than we propose in this paper, NooJ has already been tried in the area of multi word expressions, including, among others, Bulgarian verbal idioms (Todorova, 2008), Italian support verb constructions (Chatzitheodorou, 2014), English phrasal verbs (Machonis, 2010(Machonis, , 2012 and Greek idioms (Gavrilidou et al, 2012).…”