The goal of annotation standards is to ensure consistency across different corpora and languages. But do they succeed? In our paper we experiment with morphologically annotated Hungarian corpora of different sizes (ELTE DH gold standard corpus, NYTK-NerKor, and Szeged Treebank) to assess their compatibility as a merged training corpus for morphological analysis and disambiguation. Our results show that combining any two corpora not only failed to improve the results of the trained tagger but even degraded them due the inconsistent annotations. Further analysis of the annotation differences among the corpora revealed inconsistencies of several sources: different theoretical approach, lack of consensus, and tagset conversion issues.

In this article we introduce huPWKP, the first parallel corpus consisting of Hungarian standard language-simplified sentence pairs. As Hungarian is a quite low-resource language in regards to text simplification, we opted for translating an already existing corpus, PWKP (Zhu et al., 2010), on which we performed some cleaning in order to improve its quality. We evaluated the corpus both with the help of human evaluators and by training a seq2seq model on both the Hungarian corpus and the original (cleaned) English corpus. The Hungarian model performed slightly worse in terms of automatic metrics; however, the English model attains a SARI score close to the state of the art on the official PWKP set. According to the human evaluation, the corpus performs at around 3 on a scale ranging from 1 to 5 in terms of information retention and increase in simplification and around 3.7 in terms of grammaticality.

We propose a novel method for detecting optional arguments of Hungarian verbs using only positive data. We introduce a custom variant of collexeme analysis that explicitly models the noise in verb frames. Our method is, for the most part, unsupervised: we use the spectral clustering algorithm described in Brew and Schulte in Walde (2002) to build a noise model from a short, manually verified seed list of verbs. We experimented with both raw count- and context-based clusterings and found their performance almost identical. The code for our algorithm and the frame list are freely available at http://hlt.bme.hu/en/resources/tade.