Much Ado About Nothing – Identification of Zero Copulas in Hungarian Using an NMT Model

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 4802–4810 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 4802 Much Ado About Nothing Identification of Zero Copulas in Hungarian Using an NMT Model Andrea Dömötör1,2, Zijian Gy˝oz˝o Yang1, Attila Novák1 1MTA-PPKE Hungarian Language Technology Research Group, Pázmány Péter Catholic University, Faculty of Information Technology and Bionics Práter u. 50/a, 1083 Budapest, Hungary 2Pázmány Péter Catholic University, Faculty of Humanities and Social Sciences Egyetem u. 1, 2087 Piliscsaba, Hungary {surname.firstname}@itk.ppke.hu Abstract The research presented in this paper concerns zero copulas in Hungarian, i.e. the phenomenon that nominal predicates lack an explicit verbal copula in the default present tense 3rd person indicative case. We created a tool based on the state-of-the-art transformer architecture implemented in Marian NMT framework that can identify and mark the location of zero copulas, i.e. the position where an overt copula would appear in the non-default cases. Our primary aim was to support quantitative corpus-based linguistic research by creating a tool that can be used to compile a corpus of significant size containing examples of nominal predicates including the location of the zero copulas. We created the training corpus for our system transforming sentences containing overt copulas into ones containing zero copula labels. However, we first needed to disambiguate occurrences of the massively ambiguous verb van ‘exist/be/have’. We performed this using a rule-base classifier relying on English translations in the English-Hungarian parallel subcorpus of the OpenSubtitles corpus. We created several NMT-based models using different sampling methods and optionally using our baseline model to synthesize additional training data. Our best model obtains almost 90% precision and 80% recall on an in-domain test set. Keywords: zero copula, syntax, Hungarian, machine learning, transformer model, corpus linguistics, NMT 1. Introduction Zero copulas, i.e. the phenomenon that nominal predicates lack an explicit verbal copula in some default cases are fea- tured in several languages. In Hungarian, present tense 3rd person indicative is the default case. (1) a. János John is also okos clever Ø. [zerocop] ‘John is clever, too.’ b. A the disznók pigs is also okosak clever-PL Ø. [zerocop] ‘Pigs are clever, too.’ c. János John is also okos clever volt. was ‘John was clever, too.’ d. Kétlem, I_doubt hogy that János John is also okos clever lenne. would_be ‘I doubt that John would be clever, too.’ e. IÉn is also okos clever vagyok. am ‘I am clever, too.’ The purpose of our research was to create a tool that can identify zero-copular constructions and mark the location of the zero copula in sentences like (1a), i.e. the position where the copula would appear in the non-default cases (e.g. 1c). There are some special constructions in Hungarian where the presence of the verb van ‘is’ is optional. These in- clude some existential/locative constructions (2a-c) with quite strict constraints on the form of specific elements of the construction. E.g. the subject in the existential/locative construction in (2a) must include either the definite or the …

Content truncated to 500 words