Modelagem linguístico-computacional de construções de modificação circunstancial do português brasileiro

Abstract

This research aims to present the description and linguistic-computational modeling (cf. DIAS-DA-SILVA, 1996) of Constructions of Circumstantial Modification of Brazilian Portuguese in the Constructicon of Framenet Brasil (SALOMÃO, 2009), in the light of Frames Semantics (FILLMORE, 1982), and Unification-Based Construction Grammars – Berkeley Construction Grammar (KAY & FILLMORE, 1999) and SignBased Construction Grammar (SAG, 2012), to assist with the development of the m.knob chatbot. M.knob is a computational resource that works as a multilingual virtual tour guide that, through a semantic categorization mechanism, seeks to provide personalized information to the tourist, through a chatbot. In this context, the study and modeling of Circumstantial Modification Constructions may enhance M.Knob's semantic categorization system, as these constructions instantiate circumstantial information that restrict the information expressed in the main clause. Thus, the head lexical units of the main clauses evoke scenes – frames, to which a notion of circumstance (raised by the evocation of another frame, typically with a temporal or locational configuration) is added by the modifier that provides the circumstantial predicate. However, the expressed circumstance also restricts the event or action expressed in the first scene, that is, the event or action only takes place under such circumstance. Therefore, the constructional treatment of the Circumstantial Modification Constructions can help the system to identify the imposed constraint, expressed in the frame evoked by these constructions, improving the chatbot recommendation system. In addition, it will also be possible to analyze what restrictions the user usually imposes on attractions he / she plans to visit in order to understand what type of specific information is generally sought in the current tourism context. To foment this new functionality, this work brings a proposal of modeling 9 of these constructions. At the end, the necessary requirements are specified so that the modeled constructions can be integrated into the m.knob processing. We have proposed 3 hypothesis; (1) and (2), that were related with the features of the construction were confirmed by the modeling, while (3), because of some limitations of CARMA we could not confirm.