Automatic identification and multi-translatable translation of vocabulary terms with a combined approach

Automatic translation of out of vocabulary (OOV) terms has been extensively studied in the past, but multi-translatable OOV terms have received little attention. Multi-translatable OOV terms are OOV terms with some possible OOV synonyms, thus they have more than one correct translations. Traditional methods usually ignore such problem and neither identify/extract multi-translatable OOV terms nor translate them. This paper proposes a web-based OOV term translation method by utilizing a novel automatic multi-translatable OOV term identification and extraction approach. This approach integrates synonymous features and pattern matching to solve multi-translatable OOV term problems. A combined translation method is proposed for extracting translation candidates. To achieve high translation selection quality, we conducted statistical feature extraction, an artificial neural network combined with backward feature selection, and evolutionary parameter optimization is trained for selecting correct translations. Our method outperforms existing method with an accuracy of 82.61%.