AutoLearn extracts new translation relations for words and multiword expressions (MWE) of any category from bilingual texts of any size in high quality and prepares the information found as a conventional dictionary entry - with morpho-syntactic and semantic classifications and contextual use conditions. The function uses Lingenio's MT-system and analysis components as knowledge source, integrates its results into these and, by this bootstrapping approach, adapts dictionary and MT to the needs of the customer. Manual intervention is restricted to a very reduced number of difficult cases and can be carried out easily in an ergonomic graphical user interface, without need of effortful training. This is enabled by the underlying MT-architecture with rule-based core and additional statistical features. The use conditions connected to the new dictionary entries are derived from the local representation the considered word or expression is part of in the considered reference(s). They restrict the corresponding translation to similar cases so that interferences with other translations in the dictionary are avoided. A basic version of the function is already available in the current version of Lingenio's translate.
展开▼