Analysis of rule-based machine translation and neural machine translation approaches for translating portuguese to LIBRAS

In this paper, we propose a rule-based machine translator for Brazilian Portuguese to Brazilian Sign Language translation. This translator was implemented using the part of speech tagging and lemmatization techniques with implementations in Aelius and CoGrOO, respectively. Then, we developed a convolutional translator seeking to replicate the rule-based translation with a sintetic corpus. In this corpus, preprocessing techniques such as substitution of names, numbers and spelling errors by symbols were applied to improve processing. The translator were tested on two corpus, Bosque e OpenSub (extracted from a site of subtitles), of 69 and 36.858 lines respectively, and compared with translations generated by interpreters and translations generated by the application VLibras (LAVID-UFPB).