Old English morphological inflection generation with UniMorph. Assessment with a relational database and training guidelines

Resumen

The aim of this article is to assess the morphological inflection generation of Old English of the UniMorph data set. The method of this study is based on McCarthy et al.´s (2020) model of generation of putative morphological paradigms. The assessment includes inflections (morphological features and values), inflectional forms and stems. The question is also addressed of plausibility, understood as the effective attestedness of an inflectional form. The assessment tasks are carried out in a relational database specifically designed for filing and comparing the relevant data sets, including treebanks and databases of Old English lexicographical and textual sources. The overall conclusion is that the Old English UniMorph data set is consistent and robust. On the basis of the assessment, however, training guidelines of the generation model are proposed that include characters, diacritical marks, the prefix ge- in verbs, the superlative grade of adjectives, the adjectivally inflected participle and some local shortcomings.