A Study of Small Corpus-based NMT for Image-based Text Recognition

This paper presents an applied study that aims to build a Chinese and foreign language translation system for real-time applications, using image recognition and Neural Machine Translation (NMT) techniques. The image and text recognition engine development focus on possible technologies for recognition tasks of commonly used text items (Chinese, Portuguese, English) in Macau. The paper gives an overview of Artificial Intelligence (AI), Big Data Mining, and other cutting-edge technologies. The paper also develops an NMT for a limited corpus, with short training times for some translation models. It also explores deep learning problems, such as unsupervised learning, black-box models, and online learning, and shows how these challenges translate into fruitful avenues for future research.