Abstract
We generalize the word analogy task across languages, to provide a newintrinsic evaluation method for cross-lingual semantic spaces. We experimentwith six languages within different language families, including English,German, Spanish, Italian, Czech, and Croatian. State-of-the-art monolingualsemantic spaces are transformed into a shared space using dictionaries of wordtranslations. We compare several linear transformations and rank them forexperiments with monolingual (no transformation), bilingual (one semantic spaceis transformed to another), and multilingual (all semantic spaces aretransformed onto English space) versions of semantic spaces. We show thattested linear transformations preserve relationships between words (wordanalogies) and lead to impressive results. We achieve average accuracy of51.1%, 43.1%, and 38.2% for monolingual, bilingual, and multilingual semanticspaces, respectively.