📄 Still on arguments and adjuncts: the status of the indirect object and the adverbial adjunct relations in Universal Dependencies for Portuguese

Summary

The authors take up a long-standing problem in Portuguese grammar: how to distinguish the indirect object, that is, a prepositional argument of the verb, from a prepositional adverbial (adjunct). In UD, this opposition is replaced by the distinction between core and oblique. As a result, both types are usually labeled obl, and whether to single out arguments with the subtype obl:arg is left to the treebank authors for each language. The authors examine the UD guidelines and several Brazilian and Portuguese grammars. They show that the criteria found there contradict one another and do not work on real corpus data. These criteria include questions addressed to the verb, the deletion test, and reliance on intuition. Instead, the authors propose a semantic-discursive criterion. It was applied in annotating the second version of PetroGold, a gold treebank of the oil and gas domain. The new annotation changed 14.8% of the sentences in the corpus. An analysis of transitivity showed that roughly a quarter of the verb lemmas show no clear tendency toward either transitivity or intransitivity. The value of the work lies in providing an example of a consistent language-specific policy for annotating arguments in UD. Such a policy is expected to increase inter-annotator agreement and preserve useful linguistic information.

Key points

• PetroGold v2 contains 250,595 tokens in 8,949 sentences annotated according to UD. The new annotation of arguments and adjuncts changed 1,488 tokens, that is, 14.8% of the sentences.

• The annotation scheme is as follows:

  • obj — a non-prepositional direct object;
  • iobj — only a dative pronoun, when the sentence already contains a direct object;
  • obl:arg — a prepositional argument;
  • obl — a prepositional adjunct.

  Therefore, both «assistir o filme» and «assistir ao filme» are annotated as arguments.

• The authors examined the «todos juntos» corpus in the AC/DC service (Linguateca), where the verb «colocar» occurs 313,047 times. Among these occurrences there are examples without a locative adjunct, which refutes the grammarians' claim that it is obligatory.

• Together with the corpus, a table of "verb–preposition" pairs, with an example noun for each entry, has been published at petroles.puc-rio.ai.

• The transitivity statistics cover 9,653 verb occurrences and 719 lemmas, that is, 66% of all verb lemmas in the corpus. Participles and verbs with the pronoun «se» were excluded from the analysis. Most lemmas always take an argument: obj, iobj, obl:arg, xcomp or ccomp. The intermediate group between "never" and "always" comprises 25.8% of the lemmas.