Hybrid Strategies for better products and shorter time-to-market
Kurt Eberle · 2014
The main Lingenio MT products are based on rule-based architectures. In the presentation we show how knowledge from corpora is integrated into the systems using the language analysis-and translation-components in a bootstrapping approach. This relates to the bilingual dictionaries, but also to learning decisions concerning the selection of syntactic rules and semantic readings in parsing and semantic evaluation. These strategies contribute both to improve the quality of the systems and to shorten go-to-market of new products significantly. Also a number of attractive spinoff functions can be generated from them which, in addition, can be used for designing new types of products and as preparatory and postediting features in MT systems whose core is of type SMT.