Hermes: Um Modelo de Interface de Linguagem Natural para Transformação em Software

• Michael William Chagas Universidade do Vale do Rio dos Sinos (UNISINOS)
• Kleinner Farias Universidade do Vale do Rio dos Sinos (UNISINOS)
• Lucian Gonçales Universidade do Vale do Rio dos Sinos (UNISINOS)
• Lucas Kupssinskü Universidade do Vale do Rio dos Sinos (UNISINOS)
• João Carlos Gluz Instituto Federal do Rio Grande do Sul (IFRS)

Resumo

Software maintenance is a costly task and error-prone for both software developers and users as well. By knowing how and what software requirements need to be changed, end users could perform maintenance assisted by tools. However, current literature lacks for tools that support automated maintenance in real-world scenarios and allow users interaction via natural language. Even worse, the current tools are unable to understand the semantic of requests, as well as perform the necessary transformations in the maintenance software. This paper, therefore, proposes Hermes, a natural language interface model for software transformation. It combines computational linguistics techniques and logic programming to perform automated maintenance requests in software. Hermes interacts with end user through state of the art language parsers and domain ontologies by interpreting the semantics of changes requests to build a typed graph that change the software. Hermes was evaluated through an empirical study with 8 participants to investigate its performance, the level of acceptance, and usability. The collected data show that Hermes was accurate, producing a high elevated correctness number of hits by finding correct transformations and has been highly accepted by the users. The results are encouraging and show the potential for using Hermes to properly produce software maintenance requests.

Palavras-chave: Natural Language Processing, Parsing, Ontologies