Typical Sentences as a Resource for Valence

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 5276–5281 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 5276 Typical Sentences as a Resource for Valence Uwe Quasthoff, Lars Hellan, Erik Körner, Thomas Eckart, Dirk Goldhahn, Dorothee Beermann Natural Language Processing Group, University of Leipzig, Germany Department of Language and Literature, Norwegian University of Science and Technology, Norway {quasthoff, koerner, teckart, dgoldhahn}@informatik.uni-leipzig.de {lars.hellan, dorothee.beermann}@ntnu.no Abstract We describe a methodology by which verb valence information can be derived from corpora by using subcorpora of typical sentences that are constructed in a language independent manner based on frequent POS structures. The inspection of typical sentences with a fixed verb in a certain position can show the valence information directly. Using verb fingerprints, consisting of the most typical sentence patterns the verb appears in, we are able to identify standard valence patterns and compare them against a language's valence profile. With a very limited number of training data per language, valence information for other verbs can be derived as well. Based on the Norwegian valence patterns, we are able to find comparable patterns in German where typical sentences are able to express the same situation in an equivalent way, and can so allow for the construction of verb valence pairs for a bilingual dictionary. This contribution discusses this application with a focus on the Norwegian valence dictionary NorVal. Keywords: valence, valence frames, typical sentences, minimal sentences, German, English, Norwegian, corpus 1. Introduction Recent studies of sentence pattern frequencies indicate concentrations of POS-based signatures for short recurrent patterns reflecting realizations of patterns like ‘transitive’, ‘copular’ and sentence embedding, applied to corpora of German, English and Norwegian; sentences instantiating these patterns are called typical sentences in Müller et al. (2018). From a grammatical viewpoint, notions like ‘transitive’, ‘copula’ and embedded infinitive stand for valence patterns, as well as for typical POS-signatures. Given that, one can address the questions whether corpus search for typical POS-signatures might be used to identify corpus instantiations of interesting varieties of valence frames. This is not trivial, since in terms of morpho-syntactic parameters of verb valence bound items, languages like those mentioned have between 200 and 300 types of valence frames. We here present an approach which in a corpus search allows one to ‘strip’ the POS-pattern of a sentence with various types of adjuncts down to those items representing the valence frame instantiated by the verb in question. These ‘stripped’ strings can be correlated with a set of minimal sentences representing the full range of valence frames of a language, and if a match is established between the ‘reduced’ POS-signature of the sentence in question and the POS signature of the minimal sentence representing a given valence frame VF, then the actual sentence can be hypothesized as instantiating the valence frame VF. This may allow for a procedure of semi- automatic construction of valence corpora for many languages, the prerequisites being only corpus texts for the language on the one side …

Content truncated to 500 words