Identifying Multi-Word Expressions from Parallel Corpora with Kernel Methods and Crowdsourcing
Federico Sangati, Andreas van Cranenburgh, Johanna Monti · Data Archiving and Networked Services (DANS) · 2014
We investigate new ways for identifying MWEs from parallel multi-lingual corpora, based on the non-translatability property of MWEs: an MWE cannot be translated from one language to another on a word by word basis (Sag et al., 2002;Monti, 2012). MWE and Non-Translatability TARGET CASES EXCEPTIONSFixed expressions e.g., EN. by and large → IT. *da e largo.Idioms e.g., EN.Call it a day → IT. *Chiamarlo un giorno.Proverbs e.g., EN.There's no such thing as a free lunch → IT. *Non esiste una cosa come un pranzo gratuito.Phrasal verbs e.g., EN.Bring somebody down → IT. *Portare qualcuno giù.A number of MWEs can be translated literally to all other languages, such as proper names and universal proverbs.These are therefore excluded from the scope of the current work.