Function in laguage, generalization and reproductibility

Abstract

In this paper, a technique to validate the analyst's intuition and sensitivity in attributing values to a linguistic function is presented. The classification algorithm machine learning technique is applied in two datasets of polysemic phenomena in Brazilian Portuguese –diminutives and the discourse marker (eu) acho que – to test whether the criteria adopted by an analyst are consistent and can be generalized. Testing the criteria adopted by an analyst by machine learning allows us to assess their consistency and generalization potential as well as to identify when the conditions of a model are not the ones responsible for assigning the judgment.