A computational grammar for a fragment of Nheengatu

Abstract

The availability of resources for computational processing is one of the survival factors of a language. The goal of this work was to implement a fragment of Nheengatu in the Grammatical Framework formalism, specially designed for the development of multilingual applications. Once more widely spoken than Portuguese in the Amazon region, Nheengatu is threatened with extinction, although it still has an estimated number of 14,000 speakers. The fragment is restricted to sentences that express contingent and non-contingent states, but includes structurally complex grammatical phenomena typical of the Tupí-Guaraní family, which strongly contrast with the equivalent constructions in Portuguese and English. It constitutes one of the modules of GrammYEP, a multilingual computational grammar comprising equivalent English and Portuguese modules. The starting point of the implementation was the non-formalized grammatical descriptions of Navarro (2011) and Cruz (2011). The formalization revealed gaps and inconsistencies in these approaches, which were partly remedied through a reanalysis of the data. GrammYEP achieved quite satisfactory results in the translation from and to Nheengatu. It translated into Portuguese and English all examples from a test set with 142 Nheengatu sentences. Conversely, 98.18% and 84.11% of the corresponding Portuguese and English test sets were rendered into Nheengatu. On the other hand, it parsed only two examples from a negative test set with 171 ungrammatical constructions in Nheengatu. This evaluation resulted in a treebank with 243 Nheengatu sentences, paired with the equivalent sentences in Portuguese and English.