how to strongaccountstrong for idiomatic german support verb constructions.pdfVIP

How to Account for Idiomatic German Support Verb Constructions in Statistical Machine Translation 1 2 1,2 2 Fabienne Cap , Manju Nirmal , Marion Weller , Sabine Schulte im Walde 1 CIS, Ludwig-Maximilian University of Munich – cap@cis.uni-muenchen.de 2 IMS, University of Stuttgart – (nirmalmu |wellermn |schulte)@ims.uni-stuttgart.de Abstract 2003; Langer, 2009). For example, the verb take does not contribute its full meaning to the SVC take Support-verb constructions (i.e., multiword a bath, but nevertheless its semantic contribution is expressions combining a semantically light different to the verb make in the SVC make a bath verb with a predicative noun) are problem- atic for standard statistical machine translation (Butt, 2003). Often, an SVC is close in meaning systems, because SMT systems cannot distin- to a corresponding full verb, e.g., the SVC make a guish between literal and idiomatic uses of contribution is synonymous to the verb contribute. the verb. We work on the German to English Table 1 presents examples for English and for Ger- translation direction, for which the identifica- man SVCs and synonymous full verbs, where the tion of support-verb constructions is challeng- predicative nouns are embedded in a noun phrase ing due to the relatively free word order