Benefits of Data Augmentation for NMT-based Text Normalization of User-Generated Content

no code implementations • WS 2019 • Claudia Matos Veliz, Orphee De Clercq, Veronique Hoste

One of the most persistent characteristics of written user-generated content (UGC) is the use of non-standard words.

Data Augmentation Machine Translation +2

Comparing MT Approaches for Text Normalization

no code implementations • RANLP 2019 • Claudia Matos Veliz, Orphee De Clercq, Veronique Hoste

Regarding NMT, we find that the translations - or normalizations - coming out of this model are far from perfect and that for a low-resource language like Dutch adding additional training data works better than artificially augmenting the data.

Data Augmentation Language Modelling +3