Inferring a Semantic Representation of Text via Cross-Language Correlation Analysis